1 /* 2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "ci/ciUtilities.inline.hpp" 28 #include "classfile/systemDictionary.hpp" 29 #include "classfile/vmSymbols.hpp" 30 #include "compiler/compileBroker.hpp" 31 #include "compiler/compileLog.hpp" 32 #include "gc/shared/barrierSet.hpp" 33 #include "jfr/support/jfrIntrinsics.hpp" 34 #include "memory/resourceArea.hpp" 35 #include "oops/klass.inline.hpp" 36 #include "oops/objArrayKlass.hpp" 37 #include "opto/addnode.hpp" 38 #include "opto/arraycopynode.hpp" 39 #include "opto/c2compiler.hpp" 40 #include "opto/callGenerator.hpp" 41 #include "opto/castnode.hpp" 42 #include "opto/cfgnode.hpp" 43 #include "opto/convertnode.hpp" 44 #include "opto/countbitsnode.hpp" 45 #include "opto/intrinsicnode.hpp" 46 #include "opto/idealKit.hpp" 47 #include "opto/mathexactnode.hpp" 48 #include "opto/movenode.hpp" 49 #include "opto/mulnode.hpp" 50 #include "opto/narrowptrnode.hpp" 51 #include "opto/opaquenode.hpp" 52 #include "opto/parse.hpp" 53 #include "opto/runtime.hpp" 54 #include "opto/rootnode.hpp" 55 #include "opto/subnode.hpp" 56 #include "prims/nativeLookup.hpp" 57 #include "prims/unsafe.hpp" 58 #include "runtime/objectMonitor.hpp" 59 #include "runtime/sharedRuntime.hpp" 60 #include "utilities/macros.hpp" 61 #include "utilities/powerOfTwo.hpp" 62 63 class LibraryIntrinsic : public InlineCallGenerator { 64 // Extend the set of intrinsics known to the runtime: 65 public: 66 private: 67 bool _is_virtual; 68 bool _does_virtual_dispatch; 69 int8_t _predicates_count; // Intrinsic is predicated by several conditions 70 int8_t _last_predicate; // Last generated predicate 71 vmIntrinsics::ID _intrinsic_id; 72 73 public: 74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id) 75 : InlineCallGenerator(m), 76 _is_virtual(is_virtual), 77 _does_virtual_dispatch(does_virtual_dispatch), 78 _predicates_count((int8_t)predicates_count), 79 _last_predicate((int8_t)-1), 80 _intrinsic_id(id) 81 { 82 } 83 virtual bool is_intrinsic() const { return true; } 84 virtual bool is_virtual() const { return _is_virtual; } 85 virtual bool is_predicated() const { return _predicates_count > 0; } 86 virtual int predicates_count() const { return _predicates_count; } 87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; } 88 virtual JVMState* generate(JVMState* jvms); 89 virtual Node* generate_predicate(JVMState* jvms, int predicate); 90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; } 91 }; 92 93 94 // Local helper class for LibraryIntrinsic: 95 class LibraryCallKit : public GraphKit { 96 private: 97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called 98 Node* _result; // the result node, if any 99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted 100 101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type); 102 103 public: 104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic) 105 : GraphKit(jvms), 106 _intrinsic(intrinsic), 107 _result(NULL) 108 { 109 // Check if this is a root compile. In that case we don't have a caller. 110 if (!jvms->has_method()) { 111 _reexecute_sp = sp(); 112 } else { 113 // Find out how many arguments the interpreter needs when deoptimizing 114 // and save the stack pointer value so it can used by uncommon_trap. 115 // We find the argument count by looking at the declared signature. 116 bool ignored_will_link; 117 ciSignature* declared_signature = NULL; 118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature); 119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci())); 120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack 121 } 122 } 123 124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; } 125 126 ciMethod* caller() const { return jvms()->method(); } 127 int bci() const { return jvms()->bci(); } 128 LibraryIntrinsic* intrinsic() const { return _intrinsic; } 129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); } 130 ciMethod* callee() const { return _intrinsic->method(); } 131 132 bool try_to_inline(int predicate); 133 Node* try_to_predicate(int predicate); 134 135 void push_result() { 136 // Push the result onto the stack. 137 if (!stopped() && result() != NULL) { 138 BasicType bt = result()->bottom_type()->basic_type(); 139 push_node(bt, result()); 140 } 141 } 142 143 private: 144 void fatal_unexpected_iid(vmIntrinsics::ID iid) { 145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)); 146 } 147 148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; } 149 void set_result(RegionNode* region, PhiNode* value); 150 Node* result() { return _result; } 151 152 virtual int reexecute_sp() { return _reexecute_sp; } 153 154 // Helper functions to inline natives 155 Node* generate_guard(Node* test, RegionNode* region, float true_prob); 156 Node* generate_slow_guard(Node* test, RegionNode* region); 157 Node* generate_fair_guard(Node* test, RegionNode* region); 158 Node* generate_negative_guard(Node* index, RegionNode* region, 159 // resulting CastII of index: 160 Node* *pos_index = NULL); 161 Node* generate_limit_guard(Node* offset, Node* subseq_length, 162 Node* array_length, 163 RegionNode* region); 164 void generate_string_range_check(Node* array, Node* offset, 165 Node* length, bool char_count); 166 Node* generate_current_thread(Node* &tls_output); 167 Node* load_mirror_from_klass(Node* klass); 168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null, 169 RegionNode* region, int null_path, 170 int offset); 171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null, 172 RegionNode* region, int null_path) { 173 int offset = java_lang_Class::klass_offset_in_bytes(); 174 return load_klass_from_mirror_common(mirror, never_see_null, 175 region, null_path, 176 offset); 177 } 178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null, 179 RegionNode* region, int null_path) { 180 int offset = java_lang_Class::array_klass_offset_in_bytes(); 181 return load_klass_from_mirror_common(mirror, never_see_null, 182 region, null_path, 183 offset); 184 } 185 Node* generate_access_flags_guard(Node* kls, 186 int modifier_mask, int modifier_bits, 187 RegionNode* region); 188 Node* generate_interface_guard(Node* kls, RegionNode* region); 189 Node* generate_array_guard(Node* kls, RegionNode* region) { 190 return generate_array_guard_common(kls, region, false, false); 191 } 192 Node* generate_non_array_guard(Node* kls, RegionNode* region) { 193 return generate_array_guard_common(kls, region, false, true); 194 } 195 Node* generate_objArray_guard(Node* kls, RegionNode* region) { 196 return generate_array_guard_common(kls, region, true, false); 197 } 198 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) { 199 return generate_array_guard_common(kls, region, true, true); 200 } 201 Node* generate_array_guard_common(Node* kls, RegionNode* region, 202 bool obj_array, bool not_array); 203 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region); 204 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id, 205 bool is_virtual = false, bool is_static = false); 206 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) { 207 return generate_method_call(method_id, false, true); 208 } 209 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) { 210 return generate_method_call(method_id, true, false); 211 } 212 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 213 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 214 215 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae); 216 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae); 217 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae); 218 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae); 219 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 220 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae); 221 bool inline_string_indexOfChar(); 222 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae); 223 bool inline_string_toBytesU(); 224 bool inline_string_getCharsU(); 225 bool inline_string_copy(bool compress); 226 bool inline_string_char_access(bool is_store); 227 Node* round_double_node(Node* n); 228 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName); 229 bool inline_math_native(vmIntrinsics::ID id); 230 bool inline_math(vmIntrinsics::ID id); 231 bool inline_double_math(vmIntrinsics::ID id); 232 template <typename OverflowOp> 233 bool inline_math_overflow(Node* arg1, Node* arg2); 234 void inline_math_mathExact(Node* math, Node* test); 235 bool inline_math_addExactI(bool is_increment); 236 bool inline_math_addExactL(bool is_increment); 237 bool inline_math_multiplyExactI(); 238 bool inline_math_multiplyExactL(); 239 bool inline_math_multiplyHigh(); 240 bool inline_math_negateExactI(); 241 bool inline_math_negateExactL(); 242 bool inline_math_subtractExactI(bool is_decrement); 243 bool inline_math_subtractExactL(bool is_decrement); 244 bool inline_min_max(vmIntrinsics::ID id); 245 bool inline_notify(vmIntrinsics::ID id); 246 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y); 247 // This returns Type::AnyPtr, RawPtr, or OopPtr. 248 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type); 249 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false); 250 251 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind; 252 DecoratorSet mo_decorator_for_access_kind(AccessKind kind); 253 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned); 254 static bool klass_needs_init_guard(Node* kls); 255 bool inline_unsafe_allocate(); 256 bool inline_unsafe_newArray(bool uninitialized); 257 bool inline_unsafe_writeback0(); 258 bool inline_unsafe_writebackSync0(bool is_pre); 259 bool inline_unsafe_copyMemory(); 260 bool inline_native_currentThread(); 261 262 bool inline_native_time_funcs(address method, const char* funcName); 263 #ifdef JFR_HAVE_INTRINSICS 264 bool inline_native_classID(); 265 bool inline_native_getEventWriter(); 266 #endif 267 bool inline_native_Class_query(vmIntrinsics::ID id); 268 bool inline_native_subtype_check(); 269 bool inline_native_getLength(); 270 bool inline_array_copyOf(bool is_copyOfRange); 271 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae); 272 bool inline_preconditions_checkIndex(); 273 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array); 274 bool inline_native_clone(bool is_virtual); 275 bool inline_native_Reflection_getCallerClass(); 276 // Helper function for inlining native object hash method 277 bool inline_native_hashcode(bool is_virtual, bool is_static); 278 bool inline_native_getClass(); 279 280 // Helper functions for inlining arraycopy 281 bool inline_arraycopy(); 282 AllocateArrayNode* tightly_coupled_allocation(Node* ptr, 283 RegionNode* slow_region); 284 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp); 285 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp, 286 uint new_idx); 287 288 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind; 289 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind); 290 bool inline_unsafe_fence(vmIntrinsics::ID id); 291 bool inline_onspinwait(); 292 bool inline_fp_conversions(vmIntrinsics::ID id); 293 bool inline_number_methods(vmIntrinsics::ID id); 294 bool inline_reference_get(); 295 bool inline_Class_cast(); 296 bool inline_aescrypt_Block(vmIntrinsics::ID id); 297 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); 298 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id); 299 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id); 300 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); 301 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting); 302 Node* inline_counterMode_AESCrypt_predicate(); 303 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); 304 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); 305 bool inline_ghash_processBlocks(); 306 bool inline_base64_encodeBlock(); 307 bool inline_sha_implCompress(vmIntrinsics::ID id); 308 bool inline_digestBase_implCompressMB(int predicate); 309 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA, 310 bool long_state, address stubAddr, const char *stubName, 311 Node* src_start, Node* ofs, Node* limit); 312 Node* get_state_from_sha_object(Node *sha_object); 313 Node* get_state_from_sha5_object(Node *sha_object); 314 Node* inline_digestBase_implCompressMB_predicate(int predicate); 315 bool inline_encodeISOArray(); 316 bool inline_updateCRC32(); 317 bool inline_updateBytesCRC32(); 318 bool inline_updateByteBufferCRC32(); 319 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class); 320 bool inline_updateBytesCRC32C(); 321 bool inline_updateDirectByteBufferCRC32C(); 322 bool inline_updateBytesAdler32(); 323 bool inline_updateByteBufferAdler32(); 324 bool inline_multiplyToLen(); 325 bool inline_hasNegatives(); 326 bool inline_squareToLen(); 327 bool inline_mulAdd(); 328 bool inline_montgomeryMultiply(); 329 bool inline_montgomerySquare(); 330 bool inline_bigIntegerShift(bool isRightShift); 331 bool inline_vectorizedMismatch(); 332 bool inline_fma(vmIntrinsics::ID id); 333 bool inline_character_compare(vmIntrinsics::ID id); 334 bool inline_fp_min_max(vmIntrinsics::ID id); 335 336 bool inline_profileBoolean(); 337 bool inline_isCompileConstant(); 338 void clear_upper_avx() { 339 #ifdef X86 340 if (UseAVX >= 2) { 341 C->set_clear_upper_avx(true); 342 } 343 #endif 344 } 345 }; 346 347 //---------------------------make_vm_intrinsic---------------------------- 348 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 349 vmIntrinsics::ID id = m->intrinsic_id(); 350 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 351 352 if (!m->is_loaded()) { 353 // Do not attempt to inline unloaded methods. 354 return NULL; 355 } 356 357 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 358 bool is_available = false; 359 360 { 361 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 362 // the compiler must transition to '_thread_in_vm' state because both 363 // methods access VM-internal data. 364 VM_ENTRY_MARK; 365 methodHandle mh(THREAD, m->get_Method()); 366 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) && 367 !C->directive()->is_intrinsic_disabled(mh) && 368 !vmIntrinsics::is_disabled_by_flags(mh); 369 370 } 371 372 if (is_available) { 373 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 374 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 375 return new LibraryIntrinsic(m, is_virtual, 376 vmIntrinsics::predicates_needed(id), 377 vmIntrinsics::does_virtual_dispatch(id), 378 (vmIntrinsics::ID) id); 379 } else { 380 return NULL; 381 } 382 } 383 384 //----------------------register_library_intrinsics----------------------- 385 // Initialize this file's data structures, for each Compile instance. 386 void Compile::register_library_intrinsics() { 387 // Nothing to do here. 388 } 389 390 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 391 LibraryCallKit kit(jvms, this); 392 Compile* C = kit.C; 393 int nodes = C->unique(); 394 #ifndef PRODUCT 395 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 396 char buf[1000]; 397 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 398 tty->print_cr("Intrinsic %s", str); 399 } 400 #endif 401 ciMethod* callee = kit.callee(); 402 const int bci = kit.bci(); 403 404 // Try to inline the intrinsic. 405 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) && 406 kit.try_to_inline(_last_predicate)) { 407 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)" 408 : "(intrinsic)"; 409 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 410 if (C->print_intrinsics() || C->print_inlining()) { 411 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 412 } 413 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 414 if (C->log()) { 415 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 416 vmIntrinsics::name_at(intrinsic_id()), 417 (is_virtual() ? " virtual='1'" : ""), 418 C->unique() - nodes); 419 } 420 // Push the result from the inlined method onto the stack. 421 kit.push_result(); 422 C->print_inlining_update(this); 423 return kit.transfer_exceptions_into_jvms(); 424 } 425 426 // The intrinsic bailed out 427 if (jvms->has_method()) { 428 // Not a root compile. 429 const char* msg; 430 if (callee->intrinsic_candidate()) { 431 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 432 } else { 433 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 434 : "failed to inline (intrinsic), method not annotated"; 435 } 436 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg); 437 if (C->print_intrinsics() || C->print_inlining()) { 438 C->print_inlining(callee, jvms->depth() - 1, bci, msg); 439 } 440 } else { 441 // Root compile 442 ResourceMark rm; 443 stringStream msg_stream; 444 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 445 vmIntrinsics::name_at(intrinsic_id()), 446 is_virtual() ? " (virtual)" : "", bci); 447 const char *msg = msg_stream.as_string(); 448 log_debug(jit, inlining)("%s", msg); 449 if (C->print_intrinsics() || C->print_inlining()) { 450 tty->print("%s", msg); 451 } 452 } 453 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 454 C->print_inlining_update(this); 455 return NULL; 456 } 457 458 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 459 LibraryCallKit kit(jvms, this); 460 Compile* C = kit.C; 461 int nodes = C->unique(); 462 _last_predicate = predicate; 463 #ifndef PRODUCT 464 assert(is_predicated() && predicate < predicates_count(), "sanity"); 465 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 466 char buf[1000]; 467 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 468 tty->print_cr("Predicate for intrinsic %s", str); 469 } 470 #endif 471 ciMethod* callee = kit.callee(); 472 const int bci = kit.bci(); 473 474 Node* slow_ctl = kit.try_to_predicate(predicate); 475 if (!kit.failing()) { 476 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 477 : "(intrinsic, predicate)"; 478 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 479 if (C->print_intrinsics() || C->print_inlining()) { 480 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 481 } 482 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 483 if (C->log()) { 484 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 485 vmIntrinsics::name_at(intrinsic_id()), 486 (is_virtual() ? " virtual='1'" : ""), 487 C->unique() - nodes); 488 } 489 return slow_ctl; // Could be NULL if the check folds. 490 } 491 492 // The intrinsic bailed out 493 if (jvms->has_method()) { 494 // Not a root compile. 495 const char* msg = "failed to generate predicate for intrinsic"; 496 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg); 497 if (C->print_intrinsics() || C->print_inlining()) { 498 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg); 499 } 500 } else { 501 // Root compile 502 ResourceMark rm; 503 stringStream msg_stream; 504 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 505 vmIntrinsics::name_at(intrinsic_id()), 506 is_virtual() ? " (virtual)" : "", bci); 507 const char *msg = msg_stream.as_string(); 508 log_debug(jit, inlining)("%s", msg); 509 if (C->print_intrinsics() || C->print_inlining()) { 510 C->print_inlining_stream()->print("%s", msg); 511 } 512 } 513 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 514 return NULL; 515 } 516 517 bool LibraryCallKit::try_to_inline(int predicate) { 518 // Handle symbolic names for otherwise undistinguished boolean switches: 519 const bool is_store = true; 520 const bool is_compress = true; 521 const bool is_static = true; 522 const bool is_volatile = true; 523 524 if (!jvms()->has_method()) { 525 // Root JVMState has a null method. 526 assert(map()->memory()->Opcode() == Op_Parm, ""); 527 // Insert the memory aliasing node 528 set_all_memory(reset_memory()); 529 } 530 assert(merged_memory(), ""); 531 532 533 switch (intrinsic_id()) { 534 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 535 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 536 case vmIntrinsics::_getClass: return inline_native_getClass(); 537 538 case vmIntrinsics::_ceil: 539 case vmIntrinsics::_floor: 540 case vmIntrinsics::_rint: 541 case vmIntrinsics::_dsin: 542 case vmIntrinsics::_dcos: 543 case vmIntrinsics::_dtan: 544 case vmIntrinsics::_dabs: 545 case vmIntrinsics::_fabs: 546 case vmIntrinsics::_iabs: 547 case vmIntrinsics::_labs: 548 case vmIntrinsics::_datan2: 549 case vmIntrinsics::_dsqrt: 550 case vmIntrinsics::_dexp: 551 case vmIntrinsics::_dlog: 552 case vmIntrinsics::_dlog10: 553 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id()); 554 555 case vmIntrinsics::_min: 556 case vmIntrinsics::_max: return inline_min_max(intrinsic_id()); 557 558 case vmIntrinsics::_notify: 559 case vmIntrinsics::_notifyAll: 560 return inline_notify(intrinsic_id()); 561 562 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 563 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 564 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 565 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 566 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 567 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 568 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 569 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 570 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 571 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 572 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 573 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 574 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 575 576 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 577 578 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 579 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 580 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 581 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 582 583 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 584 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 585 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 586 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 587 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 588 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 589 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(); 590 591 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 592 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU); 593 594 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 595 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 596 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 597 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 598 599 case vmIntrinsics::_compressStringC: 600 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 601 case vmIntrinsics::_inflateStringC: 602 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 603 604 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 605 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 606 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 607 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 608 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 609 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 610 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 611 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 612 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 613 614 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 615 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 616 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 617 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 618 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 619 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 620 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 621 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 622 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 623 624 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 625 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 626 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 627 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 628 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 629 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 630 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 631 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 632 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 633 634 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 635 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 636 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 637 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 638 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 639 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 640 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 641 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 642 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 643 644 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 645 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 646 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 647 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 648 649 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 650 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 651 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 652 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 653 654 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 655 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 656 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 657 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 658 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 659 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 660 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 661 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 662 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 663 664 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 665 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 666 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 667 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 668 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 669 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 670 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 671 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 672 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 673 674 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 675 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 676 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 677 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 678 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 679 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 680 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 681 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 682 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 683 684 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 685 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 686 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 687 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 688 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 689 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 690 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 691 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 692 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 693 694 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 695 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 696 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 697 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 698 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 699 700 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 701 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 702 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 703 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 704 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 705 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 706 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 707 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 708 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 709 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 710 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 711 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 712 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 713 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 714 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 715 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 716 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 717 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 718 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 719 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 720 721 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 722 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 723 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 724 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 725 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 726 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 727 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 728 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 729 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 730 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 731 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 732 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 733 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 734 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 735 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 736 737 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 738 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 739 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 740 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 741 742 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 743 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 744 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 745 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 746 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 747 748 case vmIntrinsics::_loadFence: 749 case vmIntrinsics::_storeFence: 750 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 751 752 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 753 754 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 755 756 #ifdef JFR_HAVE_INTRINSICS 757 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime"); 758 case vmIntrinsics::_getClassId: return inline_native_classID(); 759 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 760 #endif 761 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 762 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 763 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 764 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 765 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 766 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 767 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 768 case vmIntrinsics::_getLength: return inline_native_getLength(); 769 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 770 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 771 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 772 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 773 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(); 774 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 775 776 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 777 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 778 779 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 780 781 case vmIntrinsics::_isInstance: 782 case vmIntrinsics::_getModifiers: 783 case vmIntrinsics::_isInterface: 784 case vmIntrinsics::_isArray: 785 case vmIntrinsics::_isPrimitive: 786 case vmIntrinsics::_getSuperclass: 787 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 788 789 case vmIntrinsics::_floatToRawIntBits: 790 case vmIntrinsics::_floatToIntBits: 791 case vmIntrinsics::_intBitsToFloat: 792 case vmIntrinsics::_doubleToRawLongBits: 793 case vmIntrinsics::_doubleToLongBits: 794 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id()); 795 796 case vmIntrinsics::_numberOfLeadingZeros_i: 797 case vmIntrinsics::_numberOfLeadingZeros_l: 798 case vmIntrinsics::_numberOfTrailingZeros_i: 799 case vmIntrinsics::_numberOfTrailingZeros_l: 800 case vmIntrinsics::_bitCount_i: 801 case vmIntrinsics::_bitCount_l: 802 case vmIntrinsics::_reverseBytes_i: 803 case vmIntrinsics::_reverseBytes_l: 804 case vmIntrinsics::_reverseBytes_s: 805 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 806 807 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 808 809 case vmIntrinsics::_Reference_get: return inline_reference_get(); 810 811 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 812 813 case vmIntrinsics::_aescrypt_encryptBlock: 814 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 815 816 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 817 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 818 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 819 820 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 821 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 822 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 823 824 case vmIntrinsics::_counterMode_AESCrypt: 825 return inline_counterMode_AESCrypt(intrinsic_id()); 826 827 case vmIntrinsics::_sha_implCompress: 828 case vmIntrinsics::_sha2_implCompress: 829 case vmIntrinsics::_sha5_implCompress: 830 return inline_sha_implCompress(intrinsic_id()); 831 832 case vmIntrinsics::_digestBase_implCompressMB: 833 return inline_digestBase_implCompressMB(predicate); 834 835 case vmIntrinsics::_multiplyToLen: 836 return inline_multiplyToLen(); 837 838 case vmIntrinsics::_squareToLen: 839 return inline_squareToLen(); 840 841 case vmIntrinsics::_mulAdd: 842 return inline_mulAdd(); 843 844 case vmIntrinsics::_montgomeryMultiply: 845 return inline_montgomeryMultiply(); 846 case vmIntrinsics::_montgomerySquare: 847 return inline_montgomerySquare(); 848 849 case vmIntrinsics::_bigIntegerRightShiftWorker: 850 return inline_bigIntegerShift(true); 851 case vmIntrinsics::_bigIntegerLeftShiftWorker: 852 return inline_bigIntegerShift(false); 853 854 case vmIntrinsics::_vectorizedMismatch: 855 return inline_vectorizedMismatch(); 856 857 case vmIntrinsics::_ghash_processBlocks: 858 return inline_ghash_processBlocks(); 859 case vmIntrinsics::_base64_encodeBlock: 860 return inline_base64_encodeBlock(); 861 862 case vmIntrinsics::_encodeISOArray: 863 case vmIntrinsics::_encodeByteISOArray: 864 return inline_encodeISOArray(); 865 866 case vmIntrinsics::_updateCRC32: 867 return inline_updateCRC32(); 868 case vmIntrinsics::_updateBytesCRC32: 869 return inline_updateBytesCRC32(); 870 case vmIntrinsics::_updateByteBufferCRC32: 871 return inline_updateByteBufferCRC32(); 872 873 case vmIntrinsics::_updateBytesCRC32C: 874 return inline_updateBytesCRC32C(); 875 case vmIntrinsics::_updateDirectByteBufferCRC32C: 876 return inline_updateDirectByteBufferCRC32C(); 877 878 case vmIntrinsics::_updateBytesAdler32: 879 return inline_updateBytesAdler32(); 880 case vmIntrinsics::_updateByteBufferAdler32: 881 return inline_updateByteBufferAdler32(); 882 883 case vmIntrinsics::_profileBoolean: 884 return inline_profileBoolean(); 885 case vmIntrinsics::_isCompileConstant: 886 return inline_isCompileConstant(); 887 888 case vmIntrinsics::_hasNegatives: 889 return inline_hasNegatives(); 890 891 case vmIntrinsics::_fmaD: 892 case vmIntrinsics::_fmaF: 893 return inline_fma(intrinsic_id()); 894 895 case vmIntrinsics::_isDigit: 896 case vmIntrinsics::_isLowerCase: 897 case vmIntrinsics::_isUpperCase: 898 case vmIntrinsics::_isWhitespace: 899 return inline_character_compare(intrinsic_id()); 900 901 case vmIntrinsics::_maxF: 902 case vmIntrinsics::_minF: 903 case vmIntrinsics::_maxD: 904 case vmIntrinsics::_minD: 905 return inline_fp_min_max(intrinsic_id()); 906 907 default: 908 // If you get here, it may be that someone has added a new intrinsic 909 // to the list in vmSymbols.hpp without implementing it here. 910 #ifndef PRODUCT 911 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 912 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 913 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 914 } 915 #endif 916 return false; 917 } 918 } 919 920 Node* LibraryCallKit::try_to_predicate(int predicate) { 921 if (!jvms()->has_method()) { 922 // Root JVMState has a null method. 923 assert(map()->memory()->Opcode() == Op_Parm, ""); 924 // Insert the memory aliasing node 925 set_all_memory(reset_memory()); 926 } 927 assert(merged_memory(), ""); 928 929 switch (intrinsic_id()) { 930 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 931 return inline_cipherBlockChaining_AESCrypt_predicate(false); 932 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 933 return inline_cipherBlockChaining_AESCrypt_predicate(true); 934 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 935 return inline_electronicCodeBook_AESCrypt_predicate(false); 936 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 937 return inline_electronicCodeBook_AESCrypt_predicate(true); 938 case vmIntrinsics::_counterMode_AESCrypt: 939 return inline_counterMode_AESCrypt_predicate(); 940 case vmIntrinsics::_digestBase_implCompressMB: 941 return inline_digestBase_implCompressMB_predicate(predicate); 942 943 default: 944 // If you get here, it may be that someone has added a new intrinsic 945 // to the list in vmSymbols.hpp without implementing it here. 946 #ifndef PRODUCT 947 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 948 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 949 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 950 } 951 #endif 952 Node* slow_ctl = control(); 953 set_control(top()); // No fast path instrinsic 954 return slow_ctl; 955 } 956 } 957 958 //------------------------------set_result------------------------------- 959 // Helper function for finishing intrinsics. 960 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 961 record_for_igvn(region); 962 set_control(_gvn.transform(region)); 963 set_result( _gvn.transform(value)); 964 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 965 } 966 967 //------------------------------generate_guard--------------------------- 968 // Helper function for generating guarded fast-slow graph structures. 969 // The given 'test', if true, guards a slow path. If the test fails 970 // then a fast path can be taken. (We generally hope it fails.) 971 // In all cases, GraphKit::control() is updated to the fast path. 972 // The returned value represents the control for the slow path. 973 // The return value is never 'top'; it is either a valid control 974 // or NULL if it is obvious that the slow path can never be taken. 975 // Also, if region and the slow control are not NULL, the slow edge 976 // is appended to the region. 977 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 978 if (stopped()) { 979 // Already short circuited. 980 return NULL; 981 } 982 983 // Build an if node and its projections. 984 // If test is true we take the slow path, which we assume is uncommon. 985 if (_gvn.type(test) == TypeInt::ZERO) { 986 // The slow branch is never taken. No need to build this guard. 987 return NULL; 988 } 989 990 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 991 992 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 993 if (if_slow == top()) { 994 // The slow branch is never taken. No need to build this guard. 995 return NULL; 996 } 997 998 if (region != NULL) 999 region->add_req(if_slow); 1000 1001 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 1002 set_control(if_fast); 1003 1004 return if_slow; 1005 } 1006 1007 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 1008 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 1009 } 1010 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 1011 return generate_guard(test, region, PROB_FAIR); 1012 } 1013 1014 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 1015 Node* *pos_index) { 1016 if (stopped()) 1017 return NULL; // already stopped 1018 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 1019 return NULL; // index is already adequately typed 1020 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 1021 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1022 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 1023 if (is_neg != NULL && pos_index != NULL) { 1024 // Emulate effect of Parse::adjust_map_after_if. 1025 Node* ccast = new CastIINode(index, TypeInt::POS); 1026 ccast->set_req(0, control()); 1027 (*pos_index) = _gvn.transform(ccast); 1028 } 1029 return is_neg; 1030 } 1031 1032 // Make sure that 'position' is a valid limit index, in [0..length]. 1033 // There are two equivalent plans for checking this: 1034 // A. (offset + copyLength) unsigned<= arrayLength 1035 // B. offset <= (arrayLength - copyLength) 1036 // We require that all of the values above, except for the sum and 1037 // difference, are already known to be non-negative. 1038 // Plan A is robust in the face of overflow, if offset and copyLength 1039 // are both hugely positive. 1040 // 1041 // Plan B is less direct and intuitive, but it does not overflow at 1042 // all, since the difference of two non-negatives is always 1043 // representable. Whenever Java methods must perform the equivalent 1044 // check they generally use Plan B instead of Plan A. 1045 // For the moment we use Plan A. 1046 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 1047 Node* subseq_length, 1048 Node* array_length, 1049 RegionNode* region) { 1050 if (stopped()) 1051 return NULL; // already stopped 1052 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 1053 if (zero_offset && subseq_length->eqv_uncast(array_length)) 1054 return NULL; // common case of whole-array copy 1055 Node* last = subseq_length; 1056 if (!zero_offset) // last += offset 1057 last = _gvn.transform(new AddINode(last, offset)); 1058 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 1059 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1060 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 1061 return is_over; 1062 } 1063 1064 // Emit range checks for the given String.value byte array 1065 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 1066 if (stopped()) { 1067 return; // already stopped 1068 } 1069 RegionNode* bailout = new RegionNode(1); 1070 record_for_igvn(bailout); 1071 if (char_count) { 1072 // Convert char count to byte count 1073 count = _gvn.transform(new LShiftINode(count, intcon(1))); 1074 } 1075 1076 // Offset and count must not be negative 1077 generate_negative_guard(offset, bailout); 1078 generate_negative_guard(count, bailout); 1079 // Offset + count must not exceed length of array 1080 generate_limit_guard(offset, count, load_array_length(array), bailout); 1081 1082 if (bailout->req() > 1) { 1083 PreserveJVMState pjvms(this); 1084 set_control(_gvn.transform(bailout)); 1085 uncommon_trap(Deoptimization::Reason_intrinsic, 1086 Deoptimization::Action_maybe_recompile); 1087 } 1088 } 1089 1090 //--------------------------generate_current_thread-------------------- 1091 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1092 ciKlass* thread_klass = env()->Thread_klass(); 1093 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 1094 Node* thread = _gvn.transform(new ThreadLocalNode()); 1095 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset())); 1096 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered)); 1097 tls_output = thread; 1098 return threadObj; 1099 } 1100 1101 1102 //------------------------------make_string_method_node------------------------ 1103 // Helper method for String intrinsic functions. This version is called with 1104 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1105 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1106 // containing the lengths of str1 and str2. 1107 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1108 Node* result = NULL; 1109 switch (opcode) { 1110 case Op_StrIndexOf: 1111 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1112 str1_start, cnt1, str2_start, cnt2, ae); 1113 break; 1114 case Op_StrComp: 1115 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1116 str1_start, cnt1, str2_start, cnt2, ae); 1117 break; 1118 case Op_StrEquals: 1119 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1120 // Use the constant length if there is one because optimized match rule may exist. 1121 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1122 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1123 break; 1124 default: 1125 ShouldNotReachHere(); 1126 return NULL; 1127 } 1128 1129 // All these intrinsics have checks. 1130 C->set_has_split_ifs(true); // Has chance for split-if optimization 1131 clear_upper_avx(); 1132 1133 return _gvn.transform(result); 1134 } 1135 1136 //------------------------------inline_string_compareTo------------------------ 1137 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1138 Node* arg1 = argument(0); 1139 Node* arg2 = argument(1); 1140 1141 arg1 = must_be_not_null(arg1, true); 1142 arg2 = must_be_not_null(arg2, true); 1143 1144 // Get start addr and length of first argument 1145 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1146 Node* arg1_cnt = load_array_length(arg1); 1147 1148 // Get start addr and length of second argument 1149 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1150 Node* arg2_cnt = load_array_length(arg2); 1151 1152 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1153 set_result(result); 1154 return true; 1155 } 1156 1157 //------------------------------inline_string_equals------------------------ 1158 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1159 Node* arg1 = argument(0); 1160 Node* arg2 = argument(1); 1161 1162 // paths (plus control) merge 1163 RegionNode* region = new RegionNode(3); 1164 Node* phi = new PhiNode(region, TypeInt::BOOL); 1165 1166 if (!stopped()) { 1167 1168 arg1 = must_be_not_null(arg1, true); 1169 arg2 = must_be_not_null(arg2, true); 1170 1171 // Get start addr and length of first argument 1172 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1173 Node* arg1_cnt = load_array_length(arg1); 1174 1175 // Get start addr and length of second argument 1176 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1177 Node* arg2_cnt = load_array_length(arg2); 1178 1179 // Check for arg1_cnt != arg2_cnt 1180 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1181 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1182 Node* if_ne = generate_slow_guard(bol, NULL); 1183 if (if_ne != NULL) { 1184 phi->init_req(2, intcon(0)); 1185 region->init_req(2, if_ne); 1186 } 1187 1188 // Check for count == 0 is done by assembler code for StrEquals. 1189 1190 if (!stopped()) { 1191 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1192 phi->init_req(1, equals); 1193 region->init_req(1, control()); 1194 } 1195 } 1196 1197 // post merge 1198 set_control(_gvn.transform(region)); 1199 record_for_igvn(region); 1200 1201 set_result(_gvn.transform(phi)); 1202 return true; 1203 } 1204 1205 //------------------------------inline_array_equals---------------------------- 1206 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1207 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1208 Node* arg1 = argument(0); 1209 Node* arg2 = argument(1); 1210 1211 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1212 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1213 clear_upper_avx(); 1214 1215 return true; 1216 } 1217 1218 //------------------------------inline_hasNegatives------------------------------ 1219 bool LibraryCallKit::inline_hasNegatives() { 1220 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1221 return false; 1222 } 1223 1224 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters"); 1225 // no receiver since it is static method 1226 Node* ba = argument(0); 1227 Node* offset = argument(1); 1228 Node* len = argument(2); 1229 1230 ba = must_be_not_null(ba, true); 1231 1232 // Range checks 1233 generate_string_range_check(ba, offset, len, false); 1234 if (stopped()) { 1235 return true; 1236 } 1237 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1238 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1239 set_result(_gvn.transform(result)); 1240 return true; 1241 } 1242 1243 bool LibraryCallKit::inline_preconditions_checkIndex() { 1244 Node* index = argument(0); 1245 Node* length = argument(1); 1246 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1247 return false; 1248 } 1249 1250 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0))); 1251 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1252 1253 { 1254 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1255 uncommon_trap(Deoptimization::Reason_intrinsic, 1256 Deoptimization::Action_make_not_entrant); 1257 } 1258 1259 if (stopped()) { 1260 return false; 1261 } 1262 1263 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length)); 1264 BoolTest::mask btest = BoolTest::lt; 1265 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1266 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1267 _gvn.set_type(rc, rc->Value(&_gvn)); 1268 if (!rc_bool->is_Con()) { 1269 record_for_igvn(rc); 1270 } 1271 set_control(_gvn.transform(new IfTrueNode(rc))); 1272 { 1273 PreserveJVMState pjvms(this); 1274 set_control(_gvn.transform(new IfFalseNode(rc))); 1275 uncommon_trap(Deoptimization::Reason_range_check, 1276 Deoptimization::Action_make_not_entrant); 1277 } 1278 1279 if (stopped()) { 1280 return false; 1281 } 1282 1283 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax)); 1284 result->set_req(0, control()); 1285 result = _gvn.transform(result); 1286 set_result(result); 1287 replace_in_map(index, result); 1288 clear_upper_avx(); 1289 return true; 1290 } 1291 1292 //------------------------------inline_string_indexOf------------------------ 1293 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1294 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1295 return false; 1296 } 1297 Node* src = argument(0); 1298 Node* tgt = argument(1); 1299 1300 // Make the merge point 1301 RegionNode* result_rgn = new RegionNode(4); 1302 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1303 1304 src = must_be_not_null(src, true); 1305 tgt = must_be_not_null(tgt, true); 1306 1307 // Get start addr and length of source string 1308 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1309 Node* src_count = load_array_length(src); 1310 1311 // Get start addr and length of substring 1312 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1313 Node* tgt_count = load_array_length(tgt); 1314 1315 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1316 // Divide src size by 2 if String is UTF16 encoded 1317 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1318 } 1319 if (ae == StrIntrinsicNode::UU) { 1320 // Divide substring size by 2 if String is UTF16 encoded 1321 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1322 } 1323 1324 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1325 if (result != NULL) { 1326 result_phi->init_req(3, result); 1327 result_rgn->init_req(3, control()); 1328 } 1329 set_control(_gvn.transform(result_rgn)); 1330 record_for_igvn(result_rgn); 1331 set_result(_gvn.transform(result_phi)); 1332 1333 return true; 1334 } 1335 1336 //-----------------------------inline_string_indexOf----------------------- 1337 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1338 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1339 return false; 1340 } 1341 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1342 return false; 1343 } 1344 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1345 Node* src = argument(0); // byte[] 1346 Node* src_count = argument(1); // char count 1347 Node* tgt = argument(2); // byte[] 1348 Node* tgt_count = argument(3); // char count 1349 Node* from_index = argument(4); // char index 1350 1351 src = must_be_not_null(src, true); 1352 tgt = must_be_not_null(tgt, true); 1353 1354 // Multiply byte array index by 2 if String is UTF16 encoded 1355 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1356 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1357 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1358 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1359 1360 // Range checks 1361 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1362 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1363 if (stopped()) { 1364 return true; 1365 } 1366 1367 RegionNode* region = new RegionNode(5); 1368 Node* phi = new PhiNode(region, TypeInt::INT); 1369 1370 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1371 if (result != NULL) { 1372 // The result is index relative to from_index if substring was found, -1 otherwise. 1373 // Generate code which will fold into cmove. 1374 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1375 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1376 1377 Node* if_lt = generate_slow_guard(bol, NULL); 1378 if (if_lt != NULL) { 1379 // result == -1 1380 phi->init_req(3, result); 1381 region->init_req(3, if_lt); 1382 } 1383 if (!stopped()) { 1384 result = _gvn.transform(new AddINode(result, from_index)); 1385 phi->init_req(4, result); 1386 region->init_req(4, control()); 1387 } 1388 } 1389 1390 set_control(_gvn.transform(region)); 1391 record_for_igvn(region); 1392 set_result(_gvn.transform(phi)); 1393 clear_upper_avx(); 1394 1395 return true; 1396 } 1397 1398 // Create StrIndexOfNode with fast path checks 1399 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1400 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1401 // Check for substr count > string count 1402 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1403 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1404 Node* if_gt = generate_slow_guard(bol, NULL); 1405 if (if_gt != NULL) { 1406 phi->init_req(1, intcon(-1)); 1407 region->init_req(1, if_gt); 1408 } 1409 if (!stopped()) { 1410 // Check for substr count == 0 1411 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1412 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1413 Node* if_zero = generate_slow_guard(bol, NULL); 1414 if (if_zero != NULL) { 1415 phi->init_req(2, intcon(0)); 1416 region->init_req(2, if_zero); 1417 } 1418 } 1419 if (!stopped()) { 1420 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1421 } 1422 return NULL; 1423 } 1424 1425 //-----------------------------inline_string_indexOfChar----------------------- 1426 bool LibraryCallKit::inline_string_indexOfChar() { 1427 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1428 return false; 1429 } 1430 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1431 return false; 1432 } 1433 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1434 Node* src = argument(0); // byte[] 1435 Node* tgt = argument(1); // tgt is int ch 1436 Node* from_index = argument(2); 1437 Node* max = argument(3); 1438 1439 src = must_be_not_null(src, true); 1440 1441 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1))); 1442 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1443 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1444 1445 // Range checks 1446 generate_string_range_check(src, src_offset, src_count, true); 1447 if (stopped()) { 1448 return true; 1449 } 1450 1451 RegionNode* region = new RegionNode(3); 1452 Node* phi = new PhiNode(region, TypeInt::INT); 1453 1454 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none); 1455 C->set_has_split_ifs(true); // Has chance for split-if optimization 1456 _gvn.transform(result); 1457 1458 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1459 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1460 1461 Node* if_lt = generate_slow_guard(bol, NULL); 1462 if (if_lt != NULL) { 1463 // result == -1 1464 phi->init_req(2, result); 1465 region->init_req(2, if_lt); 1466 } 1467 if (!stopped()) { 1468 result = _gvn.transform(new AddINode(result, from_index)); 1469 phi->init_req(1, result); 1470 region->init_req(1, control()); 1471 } 1472 set_control(_gvn.transform(region)); 1473 record_for_igvn(region); 1474 set_result(_gvn.transform(phi)); 1475 1476 return true; 1477 } 1478 //---------------------------inline_string_copy--------------------- 1479 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1480 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1481 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1482 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1483 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1484 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1485 bool LibraryCallKit::inline_string_copy(bool compress) { 1486 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1487 return false; 1488 } 1489 int nargs = 5; // 2 oops, 3 ints 1490 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1491 1492 Node* src = argument(0); 1493 Node* src_offset = argument(1); 1494 Node* dst = argument(2); 1495 Node* dst_offset = argument(3); 1496 Node* length = argument(4); 1497 1498 // Check for allocation before we add nodes that would confuse 1499 // tightly_coupled_allocation() 1500 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1501 1502 // Figure out the size and type of the elements we will be copying. 1503 const Type* src_type = src->Value(&_gvn); 1504 const Type* dst_type = dst->Value(&_gvn); 1505 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1506 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1507 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1508 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1509 "Unsupported array types for inline_string_copy"); 1510 1511 src = must_be_not_null(src, true); 1512 dst = must_be_not_null(dst, true); 1513 1514 // Convert char[] offsets to byte[] offsets 1515 bool convert_src = (compress && src_elem == T_BYTE); 1516 bool convert_dst = (!compress && dst_elem == T_BYTE); 1517 if (convert_src) { 1518 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1519 } else if (convert_dst) { 1520 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1521 } 1522 1523 // Range checks 1524 generate_string_range_check(src, src_offset, length, convert_src); 1525 generate_string_range_check(dst, dst_offset, length, convert_dst); 1526 if (stopped()) { 1527 return true; 1528 } 1529 1530 Node* src_start = array_element_address(src, src_offset, src_elem); 1531 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1532 // 'src_start' points to src array + scaled offset 1533 // 'dst_start' points to dst array + scaled offset 1534 Node* count = NULL; 1535 if (compress) { 1536 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1537 } else { 1538 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1539 } 1540 1541 if (alloc != NULL) { 1542 if (alloc->maybe_set_complete(&_gvn)) { 1543 // "You break it, you buy it." 1544 InitializeNode* init = alloc->initialization(); 1545 assert(init->is_complete(), "we just did this"); 1546 init->set_complete_with_arraycopy(); 1547 assert(dst->is_CheckCastPP(), "sanity"); 1548 assert(dst->in(0)->in(0) == init, "dest pinned"); 1549 } 1550 // Do not let stores that initialize this object be reordered with 1551 // a subsequent store that would make this object accessible by 1552 // other threads. 1553 // Record what AllocateNode this StoreStore protects so that 1554 // escape analysis can go from the MemBarStoreStoreNode to the 1555 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1556 // based on the escape status of the AllocateNode. 1557 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1558 } 1559 if (compress) { 1560 set_result(_gvn.transform(count)); 1561 } 1562 clear_upper_avx(); 1563 1564 return true; 1565 } 1566 1567 #ifdef _LP64 1568 #define XTOP ,top() /*additional argument*/ 1569 #else //_LP64 1570 #define XTOP /*no additional argument*/ 1571 #endif //_LP64 1572 1573 //------------------------inline_string_toBytesU-------------------------- 1574 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1575 bool LibraryCallKit::inline_string_toBytesU() { 1576 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1577 return false; 1578 } 1579 // Get the arguments. 1580 Node* value = argument(0); 1581 Node* offset = argument(1); 1582 Node* length = argument(2); 1583 1584 Node* newcopy = NULL; 1585 1586 // Set the original stack and the reexecute bit for the interpreter to reexecute 1587 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1588 { PreserveReexecuteState preexecs(this); 1589 jvms()->set_should_reexecute(true); 1590 1591 // Check if a null path was taken unconditionally. 1592 value = null_check(value); 1593 1594 RegionNode* bailout = new RegionNode(1); 1595 record_for_igvn(bailout); 1596 1597 // Range checks 1598 generate_negative_guard(offset, bailout); 1599 generate_negative_guard(length, bailout); 1600 generate_limit_guard(offset, length, load_array_length(value), bailout); 1601 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1602 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1603 1604 if (bailout->req() > 1) { 1605 PreserveJVMState pjvms(this); 1606 set_control(_gvn.transform(bailout)); 1607 uncommon_trap(Deoptimization::Reason_intrinsic, 1608 Deoptimization::Action_maybe_recompile); 1609 } 1610 if (stopped()) { 1611 return true; 1612 } 1613 1614 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1615 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1616 newcopy = new_array(klass_node, size, 0); // no arguments to push 1617 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL); 1618 1619 // Calculate starting addresses. 1620 Node* src_start = array_element_address(value, offset, T_CHAR); 1621 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1622 1623 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1624 const TypeInt* toffset = gvn().type(offset)->is_int(); 1625 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1626 1627 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1628 const char* copyfunc_name = "arraycopy"; 1629 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1630 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1631 OptoRuntime::fast_arraycopy_Type(), 1632 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1633 src_start, dst_start, ConvI2X(length) XTOP); 1634 // Do not let reads from the cloned object float above the arraycopy. 1635 if (alloc != NULL) { 1636 if (alloc->maybe_set_complete(&_gvn)) { 1637 // "You break it, you buy it." 1638 InitializeNode* init = alloc->initialization(); 1639 assert(init->is_complete(), "we just did this"); 1640 init->set_complete_with_arraycopy(); 1641 assert(newcopy->is_CheckCastPP(), "sanity"); 1642 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1643 } 1644 // Do not let stores that initialize this object be reordered with 1645 // a subsequent store that would make this object accessible by 1646 // other threads. 1647 // Record what AllocateNode this StoreStore protects so that 1648 // escape analysis can go from the MemBarStoreStoreNode to the 1649 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1650 // based on the escape status of the AllocateNode. 1651 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1652 } else { 1653 insert_mem_bar(Op_MemBarCPUOrder); 1654 } 1655 } // original reexecute is set back here 1656 1657 C->set_has_split_ifs(true); // Has chance for split-if optimization 1658 if (!stopped()) { 1659 set_result(newcopy); 1660 } 1661 clear_upper_avx(); 1662 1663 return true; 1664 } 1665 1666 //------------------------inline_string_getCharsU-------------------------- 1667 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1668 bool LibraryCallKit::inline_string_getCharsU() { 1669 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1670 return false; 1671 } 1672 1673 // Get the arguments. 1674 Node* src = argument(0); 1675 Node* src_begin = argument(1); 1676 Node* src_end = argument(2); // exclusive offset (i < src_end) 1677 Node* dst = argument(3); 1678 Node* dst_begin = argument(4); 1679 1680 // Check for allocation before we add nodes that would confuse 1681 // tightly_coupled_allocation() 1682 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1683 1684 // Check if a null path was taken unconditionally. 1685 src = null_check(src); 1686 dst = null_check(dst); 1687 if (stopped()) { 1688 return true; 1689 } 1690 1691 // Get length and convert char[] offset to byte[] offset 1692 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1693 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1694 1695 // Range checks 1696 generate_string_range_check(src, src_begin, length, true); 1697 generate_string_range_check(dst, dst_begin, length, false); 1698 if (stopped()) { 1699 return true; 1700 } 1701 1702 if (!stopped()) { 1703 // Calculate starting addresses. 1704 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1705 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1706 1707 // Check if array addresses are aligned to HeapWordSize 1708 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1709 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1710 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1711 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1712 1713 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1714 const char* copyfunc_name = "arraycopy"; 1715 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1716 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1717 OptoRuntime::fast_arraycopy_Type(), 1718 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1719 src_start, dst_start, ConvI2X(length) XTOP); 1720 // Do not let reads from the cloned object float above the arraycopy. 1721 if (alloc != NULL) { 1722 if (alloc->maybe_set_complete(&_gvn)) { 1723 // "You break it, you buy it." 1724 InitializeNode* init = alloc->initialization(); 1725 assert(init->is_complete(), "we just did this"); 1726 init->set_complete_with_arraycopy(); 1727 assert(dst->is_CheckCastPP(), "sanity"); 1728 assert(dst->in(0)->in(0) == init, "dest pinned"); 1729 } 1730 // Do not let stores that initialize this object be reordered with 1731 // a subsequent store that would make this object accessible by 1732 // other threads. 1733 // Record what AllocateNode this StoreStore protects so that 1734 // escape analysis can go from the MemBarStoreStoreNode to the 1735 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1736 // based on the escape status of the AllocateNode. 1737 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1738 } else { 1739 insert_mem_bar(Op_MemBarCPUOrder); 1740 } 1741 } 1742 1743 C->set_has_split_ifs(true); // Has chance for split-if optimization 1744 return true; 1745 } 1746 1747 //----------------------inline_string_char_access---------------------------- 1748 // Store/Load char to/from byte[] array. 1749 // static void StringUTF16.putChar(byte[] val, int index, int c) 1750 // static char StringUTF16.getChar(byte[] val, int index) 1751 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1752 Node* value = argument(0); 1753 Node* index = argument(1); 1754 Node* ch = is_store ? argument(2) : NULL; 1755 1756 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1757 // correctly requires matched array shapes. 1758 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1759 "sanity: byte[] and char[] bases agree"); 1760 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1761 "sanity: byte[] and char[] scales agree"); 1762 1763 // Bail when getChar over constants is requested: constant folding would 1764 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1765 // Java method would constant fold nicely instead. 1766 if (!is_store && value->is_Con() && index->is_Con()) { 1767 return false; 1768 } 1769 1770 value = must_be_not_null(value, true); 1771 1772 Node* adr = array_element_address(value, index, T_CHAR); 1773 if (adr->is_top()) { 1774 return false; 1775 } 1776 if (is_store) { 1777 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1778 } else { 1779 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 1780 set_result(ch); 1781 } 1782 return true; 1783 } 1784 1785 //--------------------------round_double_node-------------------------------- 1786 // Round a double node if necessary. 1787 Node* LibraryCallKit::round_double_node(Node* n) { 1788 if (Matcher::strict_fp_requires_explicit_rounding) { 1789 #ifdef IA32 1790 if (UseSSE < 2) { 1791 n = _gvn.transform(new RoundDoubleNode(NULL, n)); 1792 } 1793 #else 1794 Unimplemented(); 1795 #endif // IA32 1796 } 1797 return n; 1798 } 1799 1800 //------------------------------inline_math----------------------------------- 1801 // public static double Math.abs(double) 1802 // public static double Math.sqrt(double) 1803 // public static double Math.log(double) 1804 // public static double Math.log10(double) 1805 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1806 Node* arg = round_double_node(argument(0)); 1807 Node* n = NULL; 1808 switch (id) { 1809 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1810 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break; 1811 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1812 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1813 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1814 default: fatal_unexpected_iid(id); break; 1815 } 1816 set_result(_gvn.transform(n)); 1817 return true; 1818 } 1819 1820 //------------------------------inline_math----------------------------------- 1821 // public static float Math.abs(float) 1822 // public static int Math.abs(int) 1823 // public static long Math.abs(long) 1824 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1825 Node* arg = argument(0); 1826 Node* n = NULL; 1827 switch (id) { 1828 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1829 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1830 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1831 default: fatal_unexpected_iid(id); break; 1832 } 1833 set_result(_gvn.transform(n)); 1834 return true; 1835 } 1836 1837 //------------------------------runtime_math----------------------------- 1838 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1839 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1840 "must be (DD)D or (D)D type"); 1841 1842 // Inputs 1843 Node* a = round_double_node(argument(0)); 1844 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL; 1845 1846 const TypePtr* no_memory_effects = NULL; 1847 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1848 no_memory_effects, 1849 a, top(), b, b ? top() : NULL); 1850 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1851 #ifdef ASSERT 1852 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1853 assert(value_top == top(), "second value must be top"); 1854 #endif 1855 1856 set_result(value); 1857 return true; 1858 } 1859 1860 //------------------------------inline_math_native----------------------------- 1861 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1862 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f) 1863 switch (id) { 1864 // These intrinsics are not properly supported on all hardware 1865 case vmIntrinsics::_dsin: 1866 return StubRoutines::dsin() != NULL ? 1867 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1868 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN"); 1869 case vmIntrinsics::_dcos: 1870 return StubRoutines::dcos() != NULL ? 1871 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1872 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS"); 1873 case vmIntrinsics::_dtan: 1874 return StubRoutines::dtan() != NULL ? 1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1876 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN"); 1877 case vmIntrinsics::_dlog: 1878 return StubRoutines::dlog() != NULL ? 1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1880 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG"); 1881 case vmIntrinsics::_dlog10: 1882 return StubRoutines::dlog10() != NULL ? 1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1884 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10"); 1885 1886 // These intrinsics are supported on all hardware 1887 case vmIntrinsics::_ceil: 1888 case vmIntrinsics::_floor: 1889 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1890 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1891 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1892 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1893 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1894 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1895 1896 case vmIntrinsics::_dexp: 1897 return StubRoutines::dexp() != NULL ? 1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1899 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP"); 1900 case vmIntrinsics::_dpow: { 1901 Node* exp = round_double_node(argument(2)); 1902 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1903 if (d != NULL && d->getd() == 2.0) { 1904 // Special case: pow(x, 2.0) => x * x 1905 Node* base = round_double_node(argument(0)); 1906 set_result(_gvn.transform(new MulDNode(base, base))); 1907 return true; 1908 } 1909 return StubRoutines::dpow() != NULL ? 1910 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1911 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW"); 1912 } 1913 #undef FN_PTR 1914 1915 // These intrinsics are not yet correctly implemented 1916 case vmIntrinsics::_datan2: 1917 return false; 1918 1919 default: 1920 fatal_unexpected_iid(id); 1921 return false; 1922 } 1923 } 1924 1925 static bool is_simple_name(Node* n) { 1926 return (n->req() == 1 // constant 1927 || (n->is_Type() && n->as_Type()->type()->singleton()) 1928 || n->is_Proj() // parameter or return value 1929 || n->is_Phi() // local of some sort 1930 ); 1931 } 1932 1933 //----------------------------inline_notify-----------------------------------* 1934 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1935 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1936 address func; 1937 if (id == vmIntrinsics::_notify) { 1938 func = OptoRuntime::monitor_notify_Java(); 1939 } else { 1940 func = OptoRuntime::monitor_notifyAll_Java(); 1941 } 1942 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0)); 1943 make_slow_call_ex(call, env()->Throwable_klass(), false); 1944 return true; 1945 } 1946 1947 1948 //----------------------------inline_min_max----------------------------------- 1949 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1950 set_result(generate_min_max(id, argument(0), argument(1))); 1951 return true; 1952 } 1953 1954 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 1955 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 1956 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 1957 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 1958 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 1959 1960 { 1961 PreserveJVMState pjvms(this); 1962 PreserveReexecuteState preexecs(this); 1963 jvms()->set_should_reexecute(true); 1964 1965 set_control(slow_path); 1966 set_i_o(i_o()); 1967 1968 uncommon_trap(Deoptimization::Reason_intrinsic, 1969 Deoptimization::Action_none); 1970 } 1971 1972 set_control(fast_path); 1973 set_result(math); 1974 } 1975 1976 template <typename OverflowOp> 1977 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 1978 typedef typename OverflowOp::MathOp MathOp; 1979 1980 MathOp* mathOp = new MathOp(arg1, arg2); 1981 Node* operation = _gvn.transform( mathOp ); 1982 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 1983 inline_math_mathExact(operation, ofcheck); 1984 return true; 1985 } 1986 1987 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 1988 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 1989 } 1990 1991 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 1992 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 1993 } 1994 1995 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 1996 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 1997 } 1998 1999 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2000 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2001 } 2002 2003 bool LibraryCallKit::inline_math_negateExactI() { 2004 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2005 } 2006 2007 bool LibraryCallKit::inline_math_negateExactL() { 2008 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2009 } 2010 2011 bool LibraryCallKit::inline_math_multiplyExactI() { 2012 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2013 } 2014 2015 bool LibraryCallKit::inline_math_multiplyExactL() { 2016 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2017 } 2018 2019 bool LibraryCallKit::inline_math_multiplyHigh() { 2020 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2021 return true; 2022 } 2023 2024 Node* 2025 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 2026 // These are the candidate return value: 2027 Node* xvalue = x0; 2028 Node* yvalue = y0; 2029 2030 if (xvalue == yvalue) { 2031 return xvalue; 2032 } 2033 2034 bool want_max = (id == vmIntrinsics::_max); 2035 2036 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int(); 2037 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int(); 2038 if (txvalue == NULL || tyvalue == NULL) return top(); 2039 // This is not really necessary, but it is consistent with a 2040 // hypothetical MaxINode::Value method: 2041 int widen = MAX2(txvalue->_widen, tyvalue->_widen); 2042 2043 // %%% This folding logic should (ideally) be in a different place. 2044 // Some should be inside IfNode, and there to be a more reliable 2045 // transformation of ?: style patterns into cmoves. We also want 2046 // more powerful optimizations around cmove and min/max. 2047 2048 // Try to find a dominating comparison of these guys. 2049 // It can simplify the index computation for Arrays.copyOf 2050 // and similar uses of System.arraycopy. 2051 // First, compute the normalized version of CmpI(x, y). 2052 int cmp_op = Op_CmpI; 2053 Node* xkey = xvalue; 2054 Node* ykey = yvalue; 2055 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey)); 2056 if (ideal_cmpxy->is_Cmp()) { 2057 // E.g., if we have CmpI(length - offset, count), 2058 // it might idealize to CmpI(length, count + offset) 2059 cmp_op = ideal_cmpxy->Opcode(); 2060 xkey = ideal_cmpxy->in(1); 2061 ykey = ideal_cmpxy->in(2); 2062 } 2063 2064 // Start by locating any relevant comparisons. 2065 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey; 2066 Node* cmpxy = NULL; 2067 Node* cmpyx = NULL; 2068 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) { 2069 Node* cmp = start_from->fast_out(k); 2070 if (cmp->outcnt() > 0 && // must have prior uses 2071 cmp->in(0) == NULL && // must be context-independent 2072 cmp->Opcode() == cmp_op) { // right kind of compare 2073 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp; 2074 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp; 2075 } 2076 } 2077 2078 const int NCMPS = 2; 2079 Node* cmps[NCMPS] = { cmpxy, cmpyx }; 2080 int cmpn; 2081 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2082 if (cmps[cmpn] != NULL) break; // find a result 2083 } 2084 if (cmpn < NCMPS) { 2085 // Look for a dominating test that tells us the min and max. 2086 int depth = 0; // Limit search depth for speed 2087 Node* dom = control(); 2088 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) { 2089 if (++depth >= 100) break; 2090 Node* ifproj = dom; 2091 if (!ifproj->is_Proj()) continue; 2092 Node* iff = ifproj->in(0); 2093 if (!iff->is_If()) continue; 2094 Node* bol = iff->in(1); 2095 if (!bol->is_Bool()) continue; 2096 Node* cmp = bol->in(1); 2097 if (cmp == NULL) continue; 2098 for (cmpn = 0; cmpn < NCMPS; cmpn++) 2099 if (cmps[cmpn] == cmp) break; 2100 if (cmpn == NCMPS) continue; 2101 BoolTest::mask btest = bol->as_Bool()->_test._test; 2102 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate(); 2103 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2104 // At this point, we know that 'x btest y' is true. 2105 switch (btest) { 2106 case BoolTest::eq: 2107 // They are proven equal, so we can collapse the min/max. 2108 // Either value is the answer. Choose the simpler. 2109 if (is_simple_name(yvalue) && !is_simple_name(xvalue)) 2110 return yvalue; 2111 return xvalue; 2112 case BoolTest::lt: // x < y 2113 case BoolTest::le: // x <= y 2114 return (want_max ? yvalue : xvalue); 2115 case BoolTest::gt: // x > y 2116 case BoolTest::ge: // x >= y 2117 return (want_max ? xvalue : yvalue); 2118 default: 2119 break; 2120 } 2121 } 2122 } 2123 2124 // We failed to find a dominating test. 2125 // Let's pick a test that might GVN with prior tests. 2126 Node* best_bol = NULL; 2127 BoolTest::mask best_btest = BoolTest::illegal; 2128 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2129 Node* cmp = cmps[cmpn]; 2130 if (cmp == NULL) continue; 2131 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) { 2132 Node* bol = cmp->fast_out(j); 2133 if (!bol->is_Bool()) continue; 2134 BoolTest::mask btest = bol->as_Bool()->_test._test; 2135 if (btest == BoolTest::eq || btest == BoolTest::ne) continue; 2136 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2137 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) { 2138 best_bol = bol->as_Bool(); 2139 best_btest = btest; 2140 } 2141 } 2142 } 2143 2144 Node* answer_if_true = NULL; 2145 Node* answer_if_false = NULL; 2146 switch (best_btest) { 2147 default: 2148 if (cmpxy == NULL) 2149 cmpxy = ideal_cmpxy; 2150 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt)); 2151 // and fall through: 2152 case BoolTest::lt: // x < y 2153 case BoolTest::le: // x <= y 2154 answer_if_true = (want_max ? yvalue : xvalue); 2155 answer_if_false = (want_max ? xvalue : yvalue); 2156 break; 2157 case BoolTest::gt: // x > y 2158 case BoolTest::ge: // x >= y 2159 answer_if_true = (want_max ? xvalue : yvalue); 2160 answer_if_false = (want_max ? yvalue : xvalue); 2161 break; 2162 } 2163 2164 jint hi, lo; 2165 if (want_max) { 2166 // We can sharpen the minimum. 2167 hi = MAX2(txvalue->_hi, tyvalue->_hi); 2168 lo = MAX2(txvalue->_lo, tyvalue->_lo); 2169 } else { 2170 // We can sharpen the maximum. 2171 hi = MIN2(txvalue->_hi, tyvalue->_hi); 2172 lo = MIN2(txvalue->_lo, tyvalue->_lo); 2173 } 2174 2175 // Use a flow-free graph structure, to avoid creating excess control edges 2176 // which could hinder other optimizations. 2177 // Since Math.min/max is often used with arraycopy, we want 2178 // tightly_coupled_allocation to be able to see beyond min/max expressions. 2179 Node* cmov = CMoveNode::make(NULL, best_bol, 2180 answer_if_false, answer_if_true, 2181 TypeInt::make(lo, hi, widen)); 2182 2183 return _gvn.transform(cmov); 2184 2185 /* 2186 // This is not as desirable as it may seem, since Min and Max 2187 // nodes do not have a full set of optimizations. 2188 // And they would interfere, anyway, with 'if' optimizations 2189 // and with CMoveI canonical forms. 2190 switch (id) { 2191 case vmIntrinsics::_min: 2192 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break; 2193 case vmIntrinsics::_max: 2194 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break; 2195 default: 2196 ShouldNotReachHere(); 2197 } 2198 */ 2199 } 2200 2201 inline int 2202 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2203 const TypePtr* base_type = TypePtr::NULL_PTR; 2204 if (base != NULL) base_type = _gvn.type(base)->isa_ptr(); 2205 if (base_type == NULL) { 2206 // Unknown type. 2207 return Type::AnyPtr; 2208 } else if (base_type == TypePtr::NULL_PTR) { 2209 // Since this is a NULL+long form, we have to switch to a rawptr. 2210 base = _gvn.transform(new CastX2PNode(offset)); 2211 offset = MakeConX(0); 2212 return Type::RawPtr; 2213 } else if (base_type->base() == Type::RawPtr) { 2214 return Type::RawPtr; 2215 } else if (base_type->isa_oopptr()) { 2216 // Base is never null => always a heap address. 2217 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2218 return Type::OopPtr; 2219 } 2220 // Offset is small => always a heap address. 2221 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2222 if (offset_type != NULL && 2223 base_type->offset() == 0 && // (should always be?) 2224 offset_type->_lo >= 0 && 2225 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2226 return Type::OopPtr; 2227 } else if (type == T_OBJECT) { 2228 // off heap access to an oop doesn't make any sense. Has to be on 2229 // heap. 2230 return Type::OopPtr; 2231 } 2232 // Otherwise, it might either be oop+off or NULL+addr. 2233 return Type::AnyPtr; 2234 } else { 2235 // No information: 2236 return Type::AnyPtr; 2237 } 2238 } 2239 2240 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) { 2241 Node* uncasted_base = base; 2242 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2243 if (kind == Type::RawPtr) { 2244 return basic_plus_adr(top(), uncasted_base, offset); 2245 } else if (kind == Type::AnyPtr) { 2246 assert(base == uncasted_base, "unexpected base change"); 2247 if (can_cast) { 2248 if (!_gvn.type(base)->speculative_maybe_null() && 2249 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2250 // According to profiling, this access is always on 2251 // heap. Casting the base to not null and thus avoiding membars 2252 // around the access should allow better optimizations 2253 Node* null_ctl = top(); 2254 base = null_check_oop(base, &null_ctl, true, true, true); 2255 assert(null_ctl->is_top(), "no null control here"); 2256 return basic_plus_adr(base, offset); 2257 } else if (_gvn.type(base)->speculative_always_null() && 2258 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2259 // According to profiling, this access is always off 2260 // heap. 2261 base = null_assert(base); 2262 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2263 offset = MakeConX(0); 2264 return basic_plus_adr(top(), raw_base, offset); 2265 } 2266 } 2267 // We don't know if it's an on heap or off heap access. Fall back 2268 // to raw memory access. 2269 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2270 return basic_plus_adr(top(), raw, offset); 2271 } else { 2272 assert(base == uncasted_base, "unexpected base change"); 2273 // We know it's an on heap access so base can't be null 2274 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2275 base = must_be_not_null(base, true); 2276 } 2277 return basic_plus_adr(base, offset); 2278 } 2279 } 2280 2281 //--------------------------inline_number_methods----------------------------- 2282 // inline int Integer.numberOfLeadingZeros(int) 2283 // inline int Long.numberOfLeadingZeros(long) 2284 // 2285 // inline int Integer.numberOfTrailingZeros(int) 2286 // inline int Long.numberOfTrailingZeros(long) 2287 // 2288 // inline int Integer.bitCount(int) 2289 // inline int Long.bitCount(long) 2290 // 2291 // inline char Character.reverseBytes(char) 2292 // inline short Short.reverseBytes(short) 2293 // inline int Integer.reverseBytes(int) 2294 // inline long Long.reverseBytes(long) 2295 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2296 Node* arg = argument(0); 2297 Node* n = NULL; 2298 switch (id) { 2299 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2300 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2301 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2302 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2303 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2304 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2305 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2306 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2307 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2308 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2309 default: fatal_unexpected_iid(id); break; 2310 } 2311 set_result(_gvn.transform(n)); 2312 return true; 2313 } 2314 2315 //----------------------------inline_unsafe_access---------------------------- 2316 2317 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2318 // Attempt to infer a sharper value type from the offset and base type. 2319 ciKlass* sharpened_klass = NULL; 2320 2321 // See if it is an instance field, with an object type. 2322 if (alias_type->field() != NULL) { 2323 if (alias_type->field()->type()->is_klass()) { 2324 sharpened_klass = alias_type->field()->type()->as_klass(); 2325 } 2326 } 2327 2328 // See if it is a narrow oop array. 2329 if (adr_type->isa_aryptr()) { 2330 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2331 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr(); 2332 if (elem_type != NULL) { 2333 sharpened_klass = elem_type->klass(); 2334 } 2335 } 2336 } 2337 2338 // The sharpened class might be unloaded if there is no class loader 2339 // contraint in place. 2340 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) { 2341 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass); 2342 2343 #ifndef PRODUCT 2344 if (C->print_intrinsics() || C->print_inlining()) { 2345 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2346 tty->print(" sharpened value: "); tjp->dump(); tty->cr(); 2347 } 2348 #endif 2349 // Sharpen the value type. 2350 return tjp; 2351 } 2352 return NULL; 2353 } 2354 2355 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2356 switch (kind) { 2357 case Relaxed: 2358 return MO_UNORDERED; 2359 case Opaque: 2360 return MO_RELAXED; 2361 case Acquire: 2362 return MO_ACQUIRE; 2363 case Release: 2364 return MO_RELEASE; 2365 case Volatile: 2366 return MO_SEQ_CST; 2367 default: 2368 ShouldNotReachHere(); 2369 return 0; 2370 } 2371 } 2372 2373 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2374 if (callee()->is_static()) return false; // caller must have the capability! 2375 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2376 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2377 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2378 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2379 2380 if (is_reference_type(type)) { 2381 decorators |= ON_UNKNOWN_OOP_REF; 2382 } 2383 2384 if (unaligned) { 2385 decorators |= C2_UNALIGNED; 2386 } 2387 2388 #ifndef PRODUCT 2389 { 2390 ResourceMark rm; 2391 // Check the signatures. 2392 ciSignature* sig = callee()->signature(); 2393 #ifdef ASSERT 2394 if (!is_store) { 2395 // Object getReference(Object base, int/long offset), etc. 2396 BasicType rtype = sig->return_type()->basic_type(); 2397 assert(rtype == type, "getter must return the expected value"); 2398 assert(sig->count() == 2, "oop getter has 2 arguments"); 2399 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2400 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2401 } else { 2402 // void putReference(Object base, int/long offset, Object x), etc. 2403 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2404 assert(sig->count() == 3, "oop putter has 3 arguments"); 2405 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2406 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2407 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2408 assert(vtype == type, "putter must accept the expected value"); 2409 } 2410 #endif // ASSERT 2411 } 2412 #endif //PRODUCT 2413 2414 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2415 2416 Node* receiver = argument(0); // type: oop 2417 2418 // Build address expression. 2419 Node* adr; 2420 Node* heap_base_oop = top(); 2421 Node* offset = top(); 2422 Node* val; 2423 2424 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2425 Node* base = argument(1); // type: oop 2426 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2427 offset = argument(2); // type: long 2428 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2429 // to be plain byte offsets, which are also the same as those accepted 2430 // by oopDesc::field_addr. 2431 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2432 "fieldOffset must be byte-scaled"); 2433 // 32-bit machines ignore the high half! 2434 offset = ConvL2X(offset); 2435 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed); 2436 2437 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) { 2438 if (type != T_OBJECT) { 2439 decorators |= IN_NATIVE; // off-heap primitive access 2440 } else { 2441 return false; // off-heap oop accesses are not supported 2442 } 2443 } else { 2444 heap_base_oop = base; // on-heap or mixed access 2445 } 2446 2447 // Can base be NULL? Otherwise, always on-heap access. 2448 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2449 2450 if (!can_access_non_heap) { 2451 decorators |= IN_HEAP; 2452 } 2453 2454 val = is_store ? argument(4) : NULL; 2455 2456 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2457 if (adr_type == TypePtr::NULL_PTR) { 2458 return false; // off-heap access with zero address 2459 } 2460 2461 // Try to categorize the address. 2462 Compile::AliasType* alias_type = C->alias_type(adr_type); 2463 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2464 2465 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2466 alias_type->adr_type() == TypeAryPtr::RANGE) { 2467 return false; // not supported 2468 } 2469 2470 bool mismatched = false; 2471 BasicType bt = alias_type->basic_type(); 2472 if (bt != T_ILLEGAL) { 2473 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2474 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2475 // Alias type doesn't differentiate between byte[] and boolean[]). 2476 // Use address type to get the element type. 2477 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2478 } 2479 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2480 // accessing an array field with getReference is not a mismatch 2481 bt = T_OBJECT; 2482 } 2483 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2484 // Don't intrinsify mismatched object accesses 2485 return false; 2486 } 2487 mismatched = (bt != type); 2488 } else if (alias_type->adr_type()->isa_oopptr()) { 2489 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2490 } 2491 2492 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2493 2494 if (mismatched) { 2495 decorators |= C2_MISMATCHED; 2496 } 2497 2498 // First guess at the value type. 2499 const Type *value_type = Type::get_const_basic_type(type); 2500 2501 // Figure out the memory ordering. 2502 decorators |= mo_decorator_for_access_kind(kind); 2503 2504 if (!is_store && type == T_OBJECT) { 2505 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2506 if (tjp != NULL) { 2507 value_type = tjp; 2508 } 2509 } 2510 2511 receiver = null_check(receiver); 2512 if (stopped()) { 2513 return true; 2514 } 2515 // Heap pointers get a null-check from the interpreter, 2516 // as a courtesy. However, this is not guaranteed by Unsafe, 2517 // and it is not possible to fully distinguish unintended nulls 2518 // from intended ones in this API. 2519 2520 if (!is_store) { 2521 Node* p = NULL; 2522 // Try to constant fold a load from a constant field 2523 ciField* field = alias_type->field(); 2524 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) { 2525 // final or stable field 2526 p = make_constant_from_field(field, heap_base_oop); 2527 } 2528 2529 if (p == NULL) { // Could not constant fold the load 2530 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2531 // Normalize the value returned by getBoolean in the following cases 2532 if (type == T_BOOLEAN && 2533 (mismatched || 2534 heap_base_oop == top() || // - heap_base_oop is NULL or 2535 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL 2536 // and the unsafe access is made to large offset 2537 // (i.e., larger than the maximum offset necessary for any 2538 // field access) 2539 ) { 2540 IdealKit ideal = IdealKit(this); 2541 #define __ ideal. 2542 IdealVariable normalized_result(ideal); 2543 __ declarations_done(); 2544 __ set(normalized_result, p); 2545 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2546 __ set(normalized_result, ideal.ConI(1)); 2547 ideal.end_if(); 2548 final_sync(ideal); 2549 p = __ value(normalized_result); 2550 #undef __ 2551 } 2552 } 2553 if (type == T_ADDRESS) { 2554 p = gvn().transform(new CastP2XNode(NULL, p)); 2555 p = ConvX2UL(p); 2556 } 2557 // The load node has the control of the preceding MemBarCPUOrder. All 2558 // following nodes will have the control of the MemBarCPUOrder inserted at 2559 // the end of this method. So, pushing the load onto the stack at a later 2560 // point is fine. 2561 set_result(p); 2562 } else { 2563 if (bt == T_ADDRESS) { 2564 // Repackage the long as a pointer. 2565 val = ConvL2X(val); 2566 val = gvn().transform(new CastX2PNode(val)); 2567 } 2568 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2569 } 2570 2571 return true; 2572 } 2573 2574 //----------------------------inline_unsafe_load_store---------------------------- 2575 // This method serves a couple of different customers (depending on LoadStoreKind): 2576 // 2577 // LS_cmp_swap: 2578 // 2579 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2580 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2581 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2582 // 2583 // LS_cmp_swap_weak: 2584 // 2585 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2586 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2587 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2588 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2589 // 2590 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2591 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2592 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2593 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2594 // 2595 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2596 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2597 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2598 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2599 // 2600 // LS_cmp_exchange: 2601 // 2602 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2603 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2604 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2605 // 2606 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2607 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2608 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2609 // 2610 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2611 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2612 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2613 // 2614 // LS_get_add: 2615 // 2616 // int getAndAddInt( Object o, long offset, int delta) 2617 // long getAndAddLong(Object o, long offset, long delta) 2618 // 2619 // LS_get_set: 2620 // 2621 // int getAndSet(Object o, long offset, int newValue) 2622 // long getAndSet(Object o, long offset, long newValue) 2623 // Object getAndSet(Object o, long offset, Object newValue) 2624 // 2625 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2626 // This basic scheme here is the same as inline_unsafe_access, but 2627 // differs in enough details that combining them would make the code 2628 // overly confusing. (This is a true fact! I originally combined 2629 // them, but even I was confused by it!) As much code/comments as 2630 // possible are retained from inline_unsafe_access though to make 2631 // the correspondences clearer. - dl 2632 2633 if (callee()->is_static()) return false; // caller must have the capability! 2634 2635 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2636 decorators |= mo_decorator_for_access_kind(access_kind); 2637 2638 #ifndef PRODUCT 2639 BasicType rtype; 2640 { 2641 ResourceMark rm; 2642 // Check the signatures. 2643 ciSignature* sig = callee()->signature(); 2644 rtype = sig->return_type()->basic_type(); 2645 switch(kind) { 2646 case LS_get_add: 2647 case LS_get_set: { 2648 // Check the signatures. 2649 #ifdef ASSERT 2650 assert(rtype == type, "get and set must return the expected type"); 2651 assert(sig->count() == 3, "get and set has 3 arguments"); 2652 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2653 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2654 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2655 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2656 #endif // ASSERT 2657 break; 2658 } 2659 case LS_cmp_swap: 2660 case LS_cmp_swap_weak: { 2661 // Check the signatures. 2662 #ifdef ASSERT 2663 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2664 assert(sig->count() == 4, "CAS has 4 arguments"); 2665 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2666 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2667 #endif // ASSERT 2668 break; 2669 } 2670 case LS_cmp_exchange: { 2671 // Check the signatures. 2672 #ifdef ASSERT 2673 assert(rtype == type, "CAS must return the expected type"); 2674 assert(sig->count() == 4, "CAS has 4 arguments"); 2675 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2676 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2677 #endif // ASSERT 2678 break; 2679 } 2680 default: 2681 ShouldNotReachHere(); 2682 } 2683 } 2684 #endif //PRODUCT 2685 2686 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2687 2688 // Get arguments: 2689 Node* receiver = NULL; 2690 Node* base = NULL; 2691 Node* offset = NULL; 2692 Node* oldval = NULL; 2693 Node* newval = NULL; 2694 switch(kind) { 2695 case LS_cmp_swap: 2696 case LS_cmp_swap_weak: 2697 case LS_cmp_exchange: { 2698 const bool two_slot_type = type2size[type] == 2; 2699 receiver = argument(0); // type: oop 2700 base = argument(1); // type: oop 2701 offset = argument(2); // type: long 2702 oldval = argument(4); // type: oop, int, or long 2703 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2704 break; 2705 } 2706 case LS_get_add: 2707 case LS_get_set: { 2708 receiver = argument(0); // type: oop 2709 base = argument(1); // type: oop 2710 offset = argument(2); // type: long 2711 oldval = NULL; 2712 newval = argument(4); // type: oop, int, or long 2713 break; 2714 } 2715 default: 2716 ShouldNotReachHere(); 2717 } 2718 2719 // Build field offset expression. 2720 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2721 // to be plain byte offsets, which are also the same as those accepted 2722 // by oopDesc::field_addr. 2723 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2724 // 32-bit machines ignore the high half of long offsets 2725 offset = ConvL2X(offset); 2726 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false); 2727 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2728 2729 Compile::AliasType* alias_type = C->alias_type(adr_type); 2730 BasicType bt = alias_type->basic_type(); 2731 if (bt != T_ILLEGAL && 2732 (is_reference_type(bt) != (type == T_OBJECT))) { 2733 // Don't intrinsify mismatched object accesses. 2734 return false; 2735 } 2736 2737 // For CAS, unlike inline_unsafe_access, there seems no point in 2738 // trying to refine types. Just use the coarse types here. 2739 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2740 const Type *value_type = Type::get_const_basic_type(type); 2741 2742 switch (kind) { 2743 case LS_get_set: 2744 case LS_cmp_exchange: { 2745 if (type == T_OBJECT) { 2746 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2747 if (tjp != NULL) { 2748 value_type = tjp; 2749 } 2750 } 2751 break; 2752 } 2753 case LS_cmp_swap: 2754 case LS_cmp_swap_weak: 2755 case LS_get_add: 2756 break; 2757 default: 2758 ShouldNotReachHere(); 2759 } 2760 2761 // Null check receiver. 2762 receiver = null_check(receiver); 2763 if (stopped()) { 2764 return true; 2765 } 2766 2767 int alias_idx = C->get_alias_index(adr_type); 2768 2769 if (is_reference_type(type)) { 2770 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2771 2772 // Transformation of a value which could be NULL pointer (CastPP #NULL) 2773 // could be delayed during Parse (for example, in adjust_map_after_if()). 2774 // Execute transformation here to avoid barrier generation in such case. 2775 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2776 newval = _gvn.makecon(TypePtr::NULL_PTR); 2777 2778 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2779 // Refine the value to a null constant, when it is known to be null 2780 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2781 } 2782 } 2783 2784 Node* result = NULL; 2785 switch (kind) { 2786 case LS_cmp_exchange: { 2787 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2788 oldval, newval, value_type, type, decorators); 2789 break; 2790 } 2791 case LS_cmp_swap_weak: 2792 decorators |= C2_WEAK_CMPXCHG; 2793 case LS_cmp_swap: { 2794 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 2795 oldval, newval, value_type, type, decorators); 2796 break; 2797 } 2798 case LS_get_set: { 2799 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 2800 newval, value_type, type, decorators); 2801 break; 2802 } 2803 case LS_get_add: { 2804 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 2805 newval, value_type, type, decorators); 2806 break; 2807 } 2808 default: 2809 ShouldNotReachHere(); 2810 } 2811 2812 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 2813 set_result(result); 2814 return true; 2815 } 2816 2817 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 2818 // Regardless of form, don't allow previous ld/st to move down, 2819 // then issue acquire, release, or volatile mem_bar. 2820 insert_mem_bar(Op_MemBarCPUOrder); 2821 switch(id) { 2822 case vmIntrinsics::_loadFence: 2823 insert_mem_bar(Op_LoadFence); 2824 return true; 2825 case vmIntrinsics::_storeFence: 2826 insert_mem_bar(Op_StoreFence); 2827 return true; 2828 case vmIntrinsics::_fullFence: 2829 insert_mem_bar(Op_MemBarVolatile); 2830 return true; 2831 default: 2832 fatal_unexpected_iid(id); 2833 return false; 2834 } 2835 } 2836 2837 bool LibraryCallKit::inline_onspinwait() { 2838 insert_mem_bar(Op_OnSpinWait); 2839 return true; 2840 } 2841 2842 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 2843 if (!kls->is_Con()) { 2844 return true; 2845 } 2846 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr(); 2847 if (klsptr == NULL) { 2848 return true; 2849 } 2850 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass(); 2851 // don't need a guard for a klass that is already initialized 2852 return !ik->is_initialized(); 2853 } 2854 2855 //----------------------------inline_unsafe_writeback0------------------------- 2856 // public native void Unsafe.writeback0(long address) 2857 bool LibraryCallKit::inline_unsafe_writeback0() { 2858 if (!Matcher::has_match_rule(Op_CacheWB)) { 2859 return false; 2860 } 2861 #ifndef PRODUCT 2862 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 2863 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 2864 ciSignature* sig = callee()->signature(); 2865 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 2866 #endif 2867 null_check_receiver(); // null-check, then ignore 2868 Node *addr = argument(1); 2869 addr = new CastX2PNode(addr); 2870 addr = _gvn.transform(addr); 2871 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 2872 flush = _gvn.transform(flush); 2873 set_memory(flush, TypeRawPtr::BOTTOM); 2874 return true; 2875 } 2876 2877 //----------------------------inline_unsafe_writeback0------------------------- 2878 // public native void Unsafe.writeback0(long address) 2879 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 2880 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 2881 return false; 2882 } 2883 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 2884 return false; 2885 } 2886 #ifndef PRODUCT 2887 assert(Matcher::has_match_rule(Op_CacheWB), 2888 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 2889 : "found match rule for CacheWBPostSync but not CacheWB")); 2890 2891 #endif 2892 null_check_receiver(); // null-check, then ignore 2893 Node *sync; 2894 if (is_pre) { 2895 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2896 } else { 2897 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2898 } 2899 sync = _gvn.transform(sync); 2900 set_memory(sync, TypeRawPtr::BOTTOM); 2901 return true; 2902 } 2903 2904 //----------------------------inline_unsafe_allocate--------------------------- 2905 // public native Object Unsafe.allocateInstance(Class<?> cls); 2906 bool LibraryCallKit::inline_unsafe_allocate() { 2907 if (callee()->is_static()) return false; // caller must have the capability! 2908 2909 null_check_receiver(); // null-check, then ignore 2910 Node* cls = null_check(argument(1)); 2911 if (stopped()) return true; 2912 2913 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2914 kls = null_check(kls); 2915 if (stopped()) return true; // argument was like int.class 2916 2917 Node* test = NULL; 2918 if (LibraryCallKit::klass_needs_init_guard(kls)) { 2919 // Note: The argument might still be an illegal value like 2920 // Serializable.class or Object[].class. The runtime will handle it. 2921 // But we must make an explicit check for initialization. 2922 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 2923 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 2924 // can generate code to load it as unsigned byte. 2925 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 2926 Node* bits = intcon(InstanceKlass::fully_initialized); 2927 test = _gvn.transform(new SubINode(inst, bits)); 2928 // The 'test' is non-zero if we need to take a slow path. 2929 } 2930 2931 Node* obj = new_instance(kls, test); 2932 set_result(obj); 2933 return true; 2934 } 2935 2936 //------------------------inline_native_time_funcs-------------- 2937 // inline code for System.currentTimeMillis() and System.nanoTime() 2938 // these have the same type and signature 2939 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 2940 const TypeFunc* tf = OptoRuntime::void_long_Type(); 2941 const TypePtr* no_memory_effects = NULL; 2942 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 2943 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 2944 #ifdef ASSERT 2945 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 2946 assert(value_top == top(), "second value must be top"); 2947 #endif 2948 set_result(value); 2949 return true; 2950 } 2951 2952 #ifdef JFR_HAVE_INTRINSICS 2953 2954 /* 2955 * oop -> myklass 2956 * myklass->trace_id |= USED 2957 * return myklass->trace_id & ~0x3 2958 */ 2959 bool LibraryCallKit::inline_native_classID() { 2960 Node* cls = null_check(argument(0), T_OBJECT); 2961 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2962 kls = null_check(kls, T_OBJECT); 2963 2964 ByteSize offset = KLASS_TRACE_ID_OFFSET; 2965 Node* insp = basic_plus_adr(kls, in_bytes(offset)); 2966 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered); 2967 2968 Node* clsused = longcon(0x01l); // set the class bit 2969 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused)); 2970 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr(); 2971 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered); 2972 2973 #ifdef TRACE_ID_META_BITS 2974 Node* mbits = longcon(~TRACE_ID_META_BITS); 2975 tvalue = _gvn.transform(new AndLNode(tvalue, mbits)); 2976 #endif 2977 #ifdef TRACE_ID_SHIFT 2978 Node* cbits = intcon(TRACE_ID_SHIFT); 2979 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits)); 2980 #endif 2981 2982 set_result(tvalue); 2983 return true; 2984 2985 } 2986 2987 bool LibraryCallKit::inline_native_getEventWriter() { 2988 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 2989 2990 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, 2991 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 2992 2993 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 2994 2995 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) ); 2996 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) ); 2997 2998 IfNode* iff_jobj_null = 2999 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN); 3000 3001 enum { _normal_path = 1, 3002 _null_path = 2, 3003 PATH_LIMIT }; 3004 3005 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3006 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3007 3008 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null)); 3009 result_rgn->init_req(_null_path, jobj_is_null); 3010 result_val->init_req(_null_path, null()); 3011 3012 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null)); 3013 set_control(jobj_is_not_null); 3014 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT, 3015 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3016 result_rgn->init_req(_normal_path, control()); 3017 result_val->init_req(_normal_path, res); 3018 3019 set_result(result_rgn, result_val); 3020 3021 return true; 3022 } 3023 3024 #endif // JFR_HAVE_INTRINSICS 3025 3026 //------------------------inline_native_currentThread------------------ 3027 bool LibraryCallKit::inline_native_currentThread() { 3028 Node* junk = NULL; 3029 set_result(generate_current_thread(junk)); 3030 return true; 3031 } 3032 3033 //---------------------------load_mirror_from_klass---------------------------- 3034 // Given a klass oop, load its java mirror (a java.lang.Class oop). 3035 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) { 3036 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset())); 3037 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3038 // mirror = ((OopHandle)mirror)->resolve(); 3039 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE); 3040 } 3041 3042 //-----------------------load_klass_from_mirror_common------------------------- 3043 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3044 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3045 // and branch to the given path on the region. 3046 // If never_see_null, take an uncommon trap on null, so we can optimistically 3047 // compile for the non-null case. 3048 // If the region is NULL, force never_see_null = true. 3049 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3050 bool never_see_null, 3051 RegionNode* region, 3052 int null_path, 3053 int offset) { 3054 if (region == NULL) never_see_null = true; 3055 Node* p = basic_plus_adr(mirror, offset); 3056 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3057 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3058 Node* null_ctl = top(); 3059 kls = null_check_oop(kls, &null_ctl, never_see_null); 3060 if (region != NULL) { 3061 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3062 region->init_req(null_path, null_ctl); 3063 } else { 3064 assert(null_ctl == top(), "no loose ends"); 3065 } 3066 return kls; 3067 } 3068 3069 //--------------------(inline_native_Class_query helpers)--------------------- 3070 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3071 // Fall through if (mods & mask) == bits, take the guard otherwise. 3072 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3073 // Branch around if the given klass has the given modifier bit set. 3074 // Like generate_guard, adds a new path onto the region. 3075 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3076 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered); 3077 Node* mask = intcon(modifier_mask); 3078 Node* bits = intcon(modifier_bits); 3079 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3080 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3081 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3082 return generate_fair_guard(bol, region); 3083 } 3084 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3085 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3086 } 3087 3088 //-------------------------inline_native_Class_query------------------- 3089 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3090 const Type* return_type = TypeInt::BOOL; 3091 Node* prim_return_value = top(); // what happens if it's a primitive class? 3092 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3093 bool expect_prim = false; // most of these guys expect to work on refs 3094 3095 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3096 3097 Node* mirror = argument(0); 3098 Node* obj = top(); 3099 3100 switch (id) { 3101 case vmIntrinsics::_isInstance: 3102 // nothing is an instance of a primitive type 3103 prim_return_value = intcon(0); 3104 obj = argument(1); 3105 break; 3106 case vmIntrinsics::_getModifiers: 3107 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3108 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3109 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3110 break; 3111 case vmIntrinsics::_isInterface: 3112 prim_return_value = intcon(0); 3113 break; 3114 case vmIntrinsics::_isArray: 3115 prim_return_value = intcon(0); 3116 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3117 break; 3118 case vmIntrinsics::_isPrimitive: 3119 prim_return_value = intcon(1); 3120 expect_prim = true; // obviously 3121 break; 3122 case vmIntrinsics::_getSuperclass: 3123 prim_return_value = null(); 3124 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3125 break; 3126 case vmIntrinsics::_getClassAccessFlags: 3127 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3128 return_type = TypeInt::INT; // not bool! 6297094 3129 break; 3130 default: 3131 fatal_unexpected_iid(id); 3132 break; 3133 } 3134 3135 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3136 if (mirror_con == NULL) return false; // cannot happen? 3137 3138 #ifndef PRODUCT 3139 if (C->print_intrinsics() || C->print_inlining()) { 3140 ciType* k = mirror_con->java_mirror_type(); 3141 if (k) { 3142 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3143 k->print_name(); 3144 tty->cr(); 3145 } 3146 } 3147 #endif 3148 3149 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 3150 RegionNode* region = new RegionNode(PATH_LIMIT); 3151 record_for_igvn(region); 3152 PhiNode* phi = new PhiNode(region, return_type); 3153 3154 // The mirror will never be null of Reflection.getClassAccessFlags, however 3155 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 3156 // if it is. See bug 4774291. 3157 3158 // For Reflection.getClassAccessFlags(), the null check occurs in 3159 // the wrong place; see inline_unsafe_access(), above, for a similar 3160 // situation. 3161 mirror = null_check(mirror); 3162 // If mirror or obj is dead, only null-path is taken. 3163 if (stopped()) return true; 3164 3165 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 3166 3167 // Now load the mirror's klass metaobject, and null-check it. 3168 // Side-effects region with the control path if the klass is null. 3169 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 3170 // If kls is null, we have a primitive mirror. 3171 phi->init_req(_prim_path, prim_return_value); 3172 if (stopped()) { set_result(region, phi); return true; } 3173 bool safe_for_replace = (region->in(_prim_path) == top()); 3174 3175 Node* p; // handy temp 3176 Node* null_ctl; 3177 3178 // Now that we have the non-null klass, we can perform the real query. 3179 // For constant classes, the query will constant-fold in LoadNode::Value. 3180 Node* query_value = top(); 3181 switch (id) { 3182 case vmIntrinsics::_isInstance: 3183 // nothing is an instance of a primitive type 3184 query_value = gen_instanceof(obj, kls, safe_for_replace); 3185 break; 3186 3187 case vmIntrinsics::_getModifiers: 3188 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 3189 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3190 break; 3191 3192 case vmIntrinsics::_isInterface: 3193 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3194 if (generate_interface_guard(kls, region) != NULL) 3195 // A guard was added. If the guard is taken, it was an interface. 3196 phi->add_req(intcon(1)); 3197 // If we fall through, it's a plain class. 3198 query_value = intcon(0); 3199 break; 3200 3201 case vmIntrinsics::_isArray: 3202 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 3203 if (generate_array_guard(kls, region) != NULL) 3204 // A guard was added. If the guard is taken, it was an array. 3205 phi->add_req(intcon(1)); 3206 // If we fall through, it's a plain class. 3207 query_value = intcon(0); 3208 break; 3209 3210 case vmIntrinsics::_isPrimitive: 3211 query_value = intcon(0); // "normal" path produces false 3212 break; 3213 3214 case vmIntrinsics::_getSuperclass: 3215 // The rules here are somewhat unfortunate, but we can still do better 3216 // with random logic than with a JNI call. 3217 // Interfaces store null or Object as _super, but must report null. 3218 // Arrays store an intermediate super as _super, but must report Object. 3219 // Other types can report the actual _super. 3220 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3221 if (generate_interface_guard(kls, region) != NULL) 3222 // A guard was added. If the guard is taken, it was an interface. 3223 phi->add_req(null()); 3224 if (generate_array_guard(kls, region) != NULL) 3225 // A guard was added. If the guard is taken, it was an array. 3226 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 3227 // If we fall through, it's a plain class. Get its _super. 3228 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 3229 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL)); 3230 null_ctl = top(); 3231 kls = null_check_oop(kls, &null_ctl); 3232 if (null_ctl != top()) { 3233 // If the guard is taken, Object.superClass is null (both klass and mirror). 3234 region->add_req(null_ctl); 3235 phi ->add_req(null()); 3236 } 3237 if (!stopped()) { 3238 query_value = load_mirror_from_klass(kls); 3239 } 3240 break; 3241 3242 case vmIntrinsics::_getClassAccessFlags: 3243 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3244 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3245 break; 3246 3247 default: 3248 fatal_unexpected_iid(id); 3249 break; 3250 } 3251 3252 // Fall-through is the normal case of a query to a real class. 3253 phi->init_req(1, query_value); 3254 region->init_req(1, control()); 3255 3256 C->set_has_split_ifs(true); // Has chance for split-if optimization 3257 set_result(region, phi); 3258 return true; 3259 } 3260 3261 //-------------------------inline_Class_cast------------------- 3262 bool LibraryCallKit::inline_Class_cast() { 3263 Node* mirror = argument(0); // Class 3264 Node* obj = argument(1); 3265 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3266 if (mirror_con == NULL) { 3267 return false; // dead path (mirror->is_top()). 3268 } 3269 if (obj == NULL || obj->is_top()) { 3270 return false; // dead path 3271 } 3272 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 3273 3274 // First, see if Class.cast() can be folded statically. 3275 // java_mirror_type() returns non-null for compile-time Class constants. 3276 ciType* tm = mirror_con->java_mirror_type(); 3277 if (tm != NULL && tm->is_klass() && 3278 tp != NULL && tp->klass() != NULL) { 3279 if (!tp->klass()->is_loaded()) { 3280 // Don't use intrinsic when class is not loaded. 3281 return false; 3282 } else { 3283 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass()); 3284 if (static_res == Compile::SSC_always_true) { 3285 // isInstance() is true - fold the code. 3286 set_result(obj); 3287 return true; 3288 } else if (static_res == Compile::SSC_always_false) { 3289 // Don't use intrinsic, have to throw ClassCastException. 3290 // If the reference is null, the non-intrinsic bytecode will 3291 // be optimized appropriately. 3292 return false; 3293 } 3294 } 3295 } 3296 3297 // Bailout intrinsic and do normal inlining if exception path is frequent. 3298 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3299 return false; 3300 } 3301 3302 // Generate dynamic checks. 3303 // Class.cast() is java implementation of _checkcast bytecode. 3304 // Do checkcast (Parse::do_checkcast()) optimizations here. 3305 3306 mirror = null_check(mirror); 3307 // If mirror is dead, only null-path is taken. 3308 if (stopped()) { 3309 return true; 3310 } 3311 3312 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive). 3313 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT }; 3314 RegionNode* region = new RegionNode(PATH_LIMIT); 3315 record_for_igvn(region); 3316 3317 // Now load the mirror's klass metaobject, and null-check it. 3318 // If kls is null, we have a primitive mirror and 3319 // nothing is an instance of a primitive type. 3320 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 3321 3322 Node* res = top(); 3323 if (!stopped()) { 3324 Node* bad_type_ctrl = top(); 3325 // Do checkcast optimizations. 3326 res = gen_checkcast(obj, kls, &bad_type_ctrl); 3327 region->init_req(_bad_type_path, bad_type_ctrl); 3328 } 3329 if (region->in(_prim_path) != top() || 3330 region->in(_bad_type_path) != top()) { 3331 // Let Interpreter throw ClassCastException. 3332 PreserveJVMState pjvms(this); 3333 set_control(_gvn.transform(region)); 3334 uncommon_trap(Deoptimization::Reason_intrinsic, 3335 Deoptimization::Action_maybe_recompile); 3336 } 3337 if (!stopped()) { 3338 set_result(res); 3339 } 3340 return true; 3341 } 3342 3343 3344 //--------------------------inline_native_subtype_check------------------------ 3345 // This intrinsic takes the JNI calls out of the heart of 3346 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 3347 bool LibraryCallKit::inline_native_subtype_check() { 3348 // Pull both arguments off the stack. 3349 Node* args[2]; // two java.lang.Class mirrors: superc, subc 3350 args[0] = argument(0); 3351 args[1] = argument(1); 3352 Node* klasses[2]; // corresponding Klasses: superk, subk 3353 klasses[0] = klasses[1] = top(); 3354 3355 enum { 3356 // A full decision tree on {superc is prim, subc is prim}: 3357 _prim_0_path = 1, // {P,N} => false 3358 // {P,P} & superc!=subc => false 3359 _prim_same_path, // {P,P} & superc==subc => true 3360 _prim_1_path, // {N,P} => false 3361 _ref_subtype_path, // {N,N} & subtype check wins => true 3362 _both_ref_path, // {N,N} & subtype check loses => false 3363 PATH_LIMIT 3364 }; 3365 3366 RegionNode* region = new RegionNode(PATH_LIMIT); 3367 Node* phi = new PhiNode(region, TypeInt::BOOL); 3368 record_for_igvn(region); 3369 3370 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 3371 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3372 int class_klass_offset = java_lang_Class::klass_offset_in_bytes(); 3373 3374 // First null-check both mirrors and load each mirror's klass metaobject. 3375 int which_arg; 3376 for (which_arg = 0; which_arg <= 1; which_arg++) { 3377 Node* arg = args[which_arg]; 3378 arg = null_check(arg); 3379 if (stopped()) break; 3380 args[which_arg] = arg; 3381 3382 Node* p = basic_plus_adr(arg, class_klass_offset); 3383 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type); 3384 klasses[which_arg] = _gvn.transform(kls); 3385 } 3386 3387 // Having loaded both klasses, test each for null. 3388 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3389 for (which_arg = 0; which_arg <= 1; which_arg++) { 3390 Node* kls = klasses[which_arg]; 3391 Node* null_ctl = top(); 3392 kls = null_check_oop(kls, &null_ctl, never_see_null); 3393 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path); 3394 region->init_req(prim_path, null_ctl); 3395 if (stopped()) break; 3396 klasses[which_arg] = kls; 3397 } 3398 3399 if (!stopped()) { 3400 // now we have two reference types, in klasses[0..1] 3401 Node* subk = klasses[1]; // the argument to isAssignableFrom 3402 Node* superk = klasses[0]; // the receiver 3403 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 3404 // now we have a successful reference subtype check 3405 region->set_req(_ref_subtype_path, control()); 3406 } 3407 3408 // If both operands are primitive (both klasses null), then 3409 // we must return true when they are identical primitives. 3410 // It is convenient to test this after the first null klass check. 3411 set_control(region->in(_prim_0_path)); // go back to first null check 3412 if (!stopped()) { 3413 // Since superc is primitive, make a guard for the superc==subc case. 3414 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 3415 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 3416 generate_guard(bol_eq, region, PROB_FAIR); 3417 if (region->req() == PATH_LIMIT+1) { 3418 // A guard was added. If the added guard is taken, superc==subc. 3419 region->swap_edges(PATH_LIMIT, _prim_same_path); 3420 region->del_req(PATH_LIMIT); 3421 } 3422 region->set_req(_prim_0_path, control()); // Not equal after all. 3423 } 3424 3425 // these are the only paths that produce 'true': 3426 phi->set_req(_prim_same_path, intcon(1)); 3427 phi->set_req(_ref_subtype_path, intcon(1)); 3428 3429 // pull together the cases: 3430 assert(region->req() == PATH_LIMIT, "sane region"); 3431 for (uint i = 1; i < region->req(); i++) { 3432 Node* ctl = region->in(i); 3433 if (ctl == NULL || ctl == top()) { 3434 region->set_req(i, top()); 3435 phi ->set_req(i, top()); 3436 } else if (phi->in(i) == NULL) { 3437 phi->set_req(i, intcon(0)); // all other paths produce 'false' 3438 } 3439 } 3440 3441 set_control(_gvn.transform(region)); 3442 set_result(_gvn.transform(phi)); 3443 return true; 3444 } 3445 3446 //---------------------generate_array_guard_common------------------------ 3447 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, 3448 bool obj_array, bool not_array) { 3449 3450 if (stopped()) { 3451 return NULL; 3452 } 3453 3454 // If obj_array/non_array==false/false: 3455 // Branch around if the given klass is in fact an array (either obj or prim). 3456 // If obj_array/non_array==false/true: 3457 // Branch around if the given klass is not an array klass of any kind. 3458 // If obj_array/non_array==true/true: 3459 // Branch around if the kls is not an oop array (kls is int[], String, etc.) 3460 // If obj_array/non_array==true/false: 3461 // Branch around if the kls is an oop array (Object[] or subtype) 3462 // 3463 // Like generate_guard, adds a new path onto the region. 3464 jint layout_con = 0; 3465 Node* layout_val = get_layout_helper(kls, layout_con); 3466 if (layout_val == NULL) { 3467 bool query = (obj_array 3468 ? Klass::layout_helper_is_objArray(layout_con) 3469 : Klass::layout_helper_is_array(layout_con)); 3470 if (query == not_array) { 3471 return NULL; // never a branch 3472 } else { // always a branch 3473 Node* always_branch = control(); 3474 if (region != NULL) 3475 region->add_req(always_branch); 3476 set_control(top()); 3477 return always_branch; 3478 } 3479 } 3480 // Now test the correct condition. 3481 jint nval = (obj_array 3482 ? (jint)(Klass::_lh_array_tag_type_value 3483 << Klass::_lh_array_tag_shift) 3484 : Klass::_lh_neutral_value); 3485 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 3486 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array 3487 // invert the test if we are looking for a non-array 3488 if (not_array) btest = BoolTest(btest).negate(); 3489 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 3490 return generate_fair_guard(bol, region); 3491 } 3492 3493 3494 //-----------------------inline_native_newArray-------------------------- 3495 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length); 3496 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 3497 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 3498 Node* mirror; 3499 Node* count_val; 3500 if (uninitialized) { 3501 mirror = argument(1); 3502 count_val = argument(2); 3503 } else { 3504 mirror = argument(0); 3505 count_val = argument(1); 3506 } 3507 3508 mirror = null_check(mirror); 3509 // If mirror or obj is dead, only null-path is taken. 3510 if (stopped()) return true; 3511 3512 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 3513 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3514 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 3515 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3516 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3517 3518 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3519 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 3520 result_reg, _slow_path); 3521 Node* normal_ctl = control(); 3522 Node* no_array_ctl = result_reg->in(_slow_path); 3523 3524 // Generate code for the slow case. We make a call to newArray(). 3525 set_control(no_array_ctl); 3526 if (!stopped()) { 3527 // Either the input type is void.class, or else the 3528 // array klass has not yet been cached. Either the 3529 // ensuing call will throw an exception, or else it 3530 // will cache the array klass for next time. 3531 PreserveJVMState pjvms(this); 3532 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray); 3533 Node* slow_result = set_results_for_java_call(slow_call); 3534 // this->control() comes from set_results_for_java_call 3535 result_reg->set_req(_slow_path, control()); 3536 result_val->set_req(_slow_path, slow_result); 3537 result_io ->set_req(_slow_path, i_o()); 3538 result_mem->set_req(_slow_path, reset_memory()); 3539 } 3540 3541 set_control(normal_ctl); 3542 if (!stopped()) { 3543 // Normal case: The array type has been cached in the java.lang.Class. 3544 // The following call works fine even if the array type is polymorphic. 3545 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3546 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 3547 result_reg->init_req(_normal_path, control()); 3548 result_val->init_req(_normal_path, obj); 3549 result_io ->init_req(_normal_path, i_o()); 3550 result_mem->init_req(_normal_path, reset_memory()); 3551 3552 if (uninitialized) { 3553 // Mark the allocation so that zeroing is skipped 3554 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn); 3555 alloc->maybe_set_complete(&_gvn); 3556 } 3557 } 3558 3559 // Return the combined state. 3560 set_i_o( _gvn.transform(result_io) ); 3561 set_all_memory( _gvn.transform(result_mem)); 3562 3563 C->set_has_split_ifs(true); // Has chance for split-if optimization 3564 set_result(result_reg, result_val); 3565 return true; 3566 } 3567 3568 //----------------------inline_native_getLength-------------------------- 3569 // public static native int java.lang.reflect.Array.getLength(Object array); 3570 bool LibraryCallKit::inline_native_getLength() { 3571 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3572 3573 Node* array = null_check(argument(0)); 3574 // If array is dead, only null-path is taken. 3575 if (stopped()) return true; 3576 3577 // Deoptimize if it is a non-array. 3578 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL); 3579 3580 if (non_array != NULL) { 3581 PreserveJVMState pjvms(this); 3582 set_control(non_array); 3583 uncommon_trap(Deoptimization::Reason_intrinsic, 3584 Deoptimization::Action_maybe_recompile); 3585 } 3586 3587 // If control is dead, only non-array-path is taken. 3588 if (stopped()) return true; 3589 3590 // The works fine even if the array type is polymorphic. 3591 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3592 Node* result = load_array_length(array); 3593 3594 C->set_has_split_ifs(true); // Has chance for split-if optimization 3595 set_result(result); 3596 return true; 3597 } 3598 3599 //------------------------inline_array_copyOf---------------------------- 3600 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 3601 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 3602 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 3603 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3604 3605 // Get the arguments. 3606 Node* original = argument(0); 3607 Node* start = is_copyOfRange? argument(1): intcon(0); 3608 Node* end = is_copyOfRange? argument(2): argument(1); 3609 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 3610 3611 Node* newcopy = NULL; 3612 3613 // Set the original stack and the reexecute bit for the interpreter to reexecute 3614 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 3615 { PreserveReexecuteState preexecs(this); 3616 jvms()->set_should_reexecute(true); 3617 3618 array_type_mirror = null_check(array_type_mirror); 3619 original = null_check(original); 3620 3621 // Check if a null path was taken unconditionally. 3622 if (stopped()) return true; 3623 3624 Node* orig_length = load_array_length(original); 3625 3626 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0); 3627 klass_node = null_check(klass_node); 3628 3629 RegionNode* bailout = new RegionNode(1); 3630 record_for_igvn(bailout); 3631 3632 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 3633 // Bail out if that is so. 3634 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout); 3635 if (not_objArray != NULL) { 3636 // Improve the klass node's type from the new optimistic assumption: 3637 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 3638 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/); 3639 Node* cast = new CastPPNode(klass_node, akls); 3640 cast->init_req(0, control()); 3641 klass_node = _gvn.transform(cast); 3642 } 3643 3644 // Bail out if either start or end is negative. 3645 generate_negative_guard(start, bailout, &start); 3646 generate_negative_guard(end, bailout, &end); 3647 3648 Node* length = end; 3649 if (_gvn.type(start) != TypeInt::ZERO) { 3650 length = _gvn.transform(new SubINode(end, start)); 3651 } 3652 3653 // Bail out if length is negative. 3654 // Without this the new_array would throw 3655 // NegativeArraySizeException but IllegalArgumentException is what 3656 // should be thrown 3657 generate_negative_guard(length, bailout, &length); 3658 3659 if (bailout->req() > 1) { 3660 PreserveJVMState pjvms(this); 3661 set_control(_gvn.transform(bailout)); 3662 uncommon_trap(Deoptimization::Reason_intrinsic, 3663 Deoptimization::Action_maybe_recompile); 3664 } 3665 3666 if (!stopped()) { 3667 // How many elements will we copy from the original? 3668 // The answer is MinI(orig_length - start, length). 3669 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 3670 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 3671 3672 // Generate a direct call to the right arraycopy function(s). 3673 // We know the copy is disjoint but we might not know if the 3674 // oop stores need checking. 3675 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 3676 // This will fail a store-check if x contains any non-nulls. 3677 3678 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 3679 // loads/stores but it is legal only if we're sure the 3680 // Arrays.copyOf would succeed. So we need all input arguments 3681 // to the copyOf to be validated, including that the copy to the 3682 // new array won't trigger an ArrayStoreException. That subtype 3683 // check can be optimized if we know something on the type of 3684 // the input array from type speculation. 3685 if (_gvn.type(klass_node)->singleton()) { 3686 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass(); 3687 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass(); 3688 3689 int test = C->static_subtype_check(superk, subk); 3690 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 3691 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 3692 if (t_original->speculative_type() != NULL) { 3693 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 3694 } 3695 } 3696 } 3697 3698 bool validated = false; 3699 // Reason_class_check rather than Reason_intrinsic because we 3700 // want to intrinsify even if this traps. 3701 if (!too_many_traps(Deoptimization::Reason_class_check)) { 3702 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node); 3703 3704 if (not_subtype_ctrl != top()) { 3705 PreserveJVMState pjvms(this); 3706 set_control(not_subtype_ctrl); 3707 uncommon_trap(Deoptimization::Reason_class_check, 3708 Deoptimization::Action_make_not_entrant); 3709 assert(stopped(), "Should be stopped"); 3710 } 3711 validated = true; 3712 } 3713 3714 if (!stopped()) { 3715 newcopy = new_array(klass_node, length, 0); // no arguments to push 3716 3717 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false, 3718 load_object_klass(original), klass_node); 3719 if (!is_copyOfRange) { 3720 ac->set_copyof(validated); 3721 } else { 3722 ac->set_copyofrange(validated); 3723 } 3724 Node* n = _gvn.transform(ac); 3725 if (n == ac) { 3726 ac->connect_outputs(this); 3727 } else { 3728 assert(validated, "shouldn't transform if all arguments not validated"); 3729 set_all_memory(n); 3730 } 3731 } 3732 } 3733 } // original reexecute is set back here 3734 3735 C->set_has_split_ifs(true); // Has chance for split-if optimization 3736 if (!stopped()) { 3737 set_result(newcopy); 3738 } 3739 return true; 3740 } 3741 3742 3743 //----------------------generate_virtual_guard--------------------------- 3744 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 3745 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 3746 RegionNode* slow_region) { 3747 ciMethod* method = callee(); 3748 int vtable_index = method->vtable_index(); 3749 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 3750 "bad index %d", vtable_index); 3751 // Get the Method* out of the appropriate vtable entry. 3752 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 3753 vtable_index*vtableEntry::size_in_bytes() + 3754 vtableEntry::method_offset_in_bytes(); 3755 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 3756 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3757 3758 // Compare the target method with the expected method (e.g., Object.hashCode). 3759 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 3760 3761 Node* native_call = makecon(native_call_addr); 3762 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 3763 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 3764 3765 return generate_slow_guard(test_native, slow_region); 3766 } 3767 3768 //-----------------------generate_method_call---------------------------- 3769 // Use generate_method_call to make a slow-call to the real 3770 // method if the fast path fails. An alternative would be to 3771 // use a stub like OptoRuntime::slow_arraycopy_Java. 3772 // This only works for expanding the current library call, 3773 // not another intrinsic. (E.g., don't use this for making an 3774 // arraycopy call inside of the copyOf intrinsic.) 3775 CallJavaNode* 3776 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) { 3777 // When compiling the intrinsic method itself, do not use this technique. 3778 guarantee(callee() != C->method(), "cannot make slow-call to self"); 3779 3780 ciMethod* method = callee(); 3781 // ensure the JVMS we have will be correct for this call 3782 guarantee(method_id == method->intrinsic_id(), "must match"); 3783 3784 const TypeFunc* tf = TypeFunc::make(method); 3785 CallJavaNode* slow_call; 3786 if (is_static) { 3787 assert(!is_virtual, ""); 3788 slow_call = new CallStaticJavaNode(C, tf, 3789 SharedRuntime::get_resolve_static_call_stub(), 3790 method, bci()); 3791 } else if (is_virtual) { 3792 null_check_receiver(); 3793 int vtable_index = Method::invalid_vtable_index; 3794 if (UseInlineCaches) { 3795 // Suppress the vtable call 3796 } else { 3797 // hashCode and clone are not a miranda methods, 3798 // so the vtable index is fixed. 3799 // No need to use the linkResolver to get it. 3800 vtable_index = method->vtable_index(); 3801 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 3802 "bad index %d", vtable_index); 3803 } 3804 slow_call = new CallDynamicJavaNode(tf, 3805 SharedRuntime::get_resolve_virtual_call_stub(), 3806 method, vtable_index, bci()); 3807 } else { // neither virtual nor static: opt_virtual 3808 null_check_receiver(); 3809 slow_call = new CallStaticJavaNode(C, tf, 3810 SharedRuntime::get_resolve_opt_virtual_call_stub(), 3811 method, bci()); 3812 slow_call->set_optimized_virtual(true); 3813 } 3814 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 3815 // To be able to issue a direct call (optimized virtual or virtual) 3816 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 3817 // about the method being invoked should be attached to the call site to 3818 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 3819 slow_call->set_override_symbolic_info(true); 3820 } 3821 set_arguments_for_java_call(slow_call); 3822 set_edges_for_java_call(slow_call); 3823 return slow_call; 3824 } 3825 3826 3827 /** 3828 * Build special case code for calls to hashCode on an object. This call may 3829 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 3830 * slightly different code. 3831 */ 3832 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 3833 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 3834 assert(!(is_virtual && is_static), "either virtual, special, or static"); 3835 3836 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 3837 3838 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3839 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 3840 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3841 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3842 Node* obj = NULL; 3843 if (!is_static) { 3844 // Check for hashing null object 3845 obj = null_check_receiver(); 3846 if (stopped()) return true; // unconditionally null 3847 result_reg->init_req(_null_path, top()); 3848 result_val->init_req(_null_path, top()); 3849 } else { 3850 // Do a null check, and return zero if null. 3851 // System.identityHashCode(null) == 0 3852 obj = argument(0); 3853 Node* null_ctl = top(); 3854 obj = null_check_oop(obj, &null_ctl); 3855 result_reg->init_req(_null_path, null_ctl); 3856 result_val->init_req(_null_path, _gvn.intcon(0)); 3857 } 3858 3859 // Unconditionally null? Then return right away. 3860 if (stopped()) { 3861 set_control( result_reg->in(_null_path)); 3862 if (!stopped()) 3863 set_result(result_val->in(_null_path)); 3864 return true; 3865 } 3866 3867 // We only go to the fast case code if we pass a number of guards. The 3868 // paths which do not pass are accumulated in the slow_region. 3869 RegionNode* slow_region = new RegionNode(1); 3870 record_for_igvn(slow_region); 3871 3872 // If this is a virtual call, we generate a funny guard. We pull out 3873 // the vtable entry corresponding to hashCode() from the target object. 3874 // If the target method which we are calling happens to be the native 3875 // Object hashCode() method, we pass the guard. We do not need this 3876 // guard for non-virtual calls -- the caller is known to be the native 3877 // Object hashCode(). 3878 if (is_virtual) { 3879 // After null check, get the object's klass. 3880 Node* obj_klass = load_object_klass(obj); 3881 generate_virtual_guard(obj_klass, slow_region); 3882 } 3883 3884 // Get the header out of the object, use LoadMarkNode when available 3885 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 3886 // The control of the load must be NULL. Otherwise, the load can move before 3887 // the null check after castPP removal. 3888 Node* no_ctrl = NULL; 3889 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 3890 3891 // Test the header to see if it is unlocked. 3892 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place); 3893 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 3894 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 3895 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val)); 3896 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne)); 3897 3898 generate_slow_guard(test_unlocked, slow_region); 3899 3900 // Get the hash value and check to see that it has been properly assigned. 3901 // We depend on hash_mask being at most 32 bits and avoid the use of 3902 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 3903 // vm: see markWord.hpp. 3904 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 3905 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 3906 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 3907 // This hack lets the hash bits live anywhere in the mark object now, as long 3908 // as the shift drops the relevant bits into the low 32 bits. Note that 3909 // Java spec says that HashCode is an int so there's no point in capturing 3910 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 3911 hshifted_header = ConvX2I(hshifted_header); 3912 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 3913 3914 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 3915 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 3916 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 3917 3918 generate_slow_guard(test_assigned, slow_region); 3919 3920 Node* init_mem = reset_memory(); 3921 // fill in the rest of the null path: 3922 result_io ->init_req(_null_path, i_o()); 3923 result_mem->init_req(_null_path, init_mem); 3924 3925 result_val->init_req(_fast_path, hash_val); 3926 result_reg->init_req(_fast_path, control()); 3927 result_io ->init_req(_fast_path, i_o()); 3928 result_mem->init_req(_fast_path, init_mem); 3929 3930 // Generate code for the slow case. We make a call to hashCode(). 3931 set_control(_gvn.transform(slow_region)); 3932 if (!stopped()) { 3933 // No need for PreserveJVMState, because we're using up the present state. 3934 set_all_memory(init_mem); 3935 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 3936 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static); 3937 Node* slow_result = set_results_for_java_call(slow_call); 3938 // this->control() comes from set_results_for_java_call 3939 result_reg->init_req(_slow_path, control()); 3940 result_val->init_req(_slow_path, slow_result); 3941 result_io ->set_req(_slow_path, i_o()); 3942 result_mem ->set_req(_slow_path, reset_memory()); 3943 } 3944 3945 // Return the combined state. 3946 set_i_o( _gvn.transform(result_io) ); 3947 set_all_memory( _gvn.transform(result_mem)); 3948 3949 set_result(result_reg, result_val); 3950 return true; 3951 } 3952 3953 //---------------------------inline_native_getClass---------------------------- 3954 // public final native Class<?> java.lang.Object.getClass(); 3955 // 3956 // Build special case code for calls to getClass on an object. 3957 bool LibraryCallKit::inline_native_getClass() { 3958 Node* obj = null_check_receiver(); 3959 if (stopped()) return true; 3960 set_result(load_mirror_from_klass(load_object_klass(obj))); 3961 return true; 3962 } 3963 3964 //-----------------inline_native_Reflection_getCallerClass--------------------- 3965 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 3966 // 3967 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 3968 // 3969 // NOTE: This code must perform the same logic as JVM_GetCallerClass 3970 // in that it must skip particular security frames and checks for 3971 // caller sensitive methods. 3972 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 3973 #ifndef PRODUCT 3974 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 3975 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 3976 } 3977 #endif 3978 3979 if (!jvms()->has_method()) { 3980 #ifndef PRODUCT 3981 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 3982 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 3983 } 3984 #endif 3985 return false; 3986 } 3987 3988 // Walk back up the JVM state to find the caller at the required 3989 // depth. 3990 JVMState* caller_jvms = jvms(); 3991 3992 // Cf. JVM_GetCallerClass 3993 // NOTE: Start the loop at depth 1 because the current JVM state does 3994 // not include the Reflection.getCallerClass() frame. 3995 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { 3996 ciMethod* m = caller_jvms->method(); 3997 switch (n) { 3998 case 0: 3999 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4000 break; 4001 case 1: 4002 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4003 if (!m->caller_sensitive()) { 4004 #ifndef PRODUCT 4005 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4006 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4007 } 4008 #endif 4009 return false; // bail-out; let JVM_GetCallerClass do the work 4010 } 4011 break; 4012 default: 4013 if (!m->is_ignored_by_security_stack_walk()) { 4014 // We have reached the desired frame; return the holder class. 4015 // Acquire method holder as java.lang.Class and push as constant. 4016 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4017 ciInstance* caller_mirror = caller_klass->java_mirror(); 4018 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4019 4020 #ifndef PRODUCT 4021 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4022 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth()); 4023 tty->print_cr(" JVM state at this point:"); 4024 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4025 ciMethod* m = jvms()->of_depth(i)->method(); 4026 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4027 } 4028 } 4029 #endif 4030 return true; 4031 } 4032 break; 4033 } 4034 } 4035 4036 #ifndef PRODUCT 4037 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4038 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4039 tty->print_cr(" JVM state at this point:"); 4040 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4041 ciMethod* m = jvms()->of_depth(i)->method(); 4042 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4043 } 4044 } 4045 #endif 4046 4047 return false; // bail-out; let JVM_GetCallerClass do the work 4048 } 4049 4050 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4051 Node* arg = argument(0); 4052 Node* result = NULL; 4053 4054 switch (id) { 4055 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4056 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4057 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4058 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4059 4060 case vmIntrinsics::_doubleToLongBits: { 4061 // two paths (plus control) merge in a wood 4062 RegionNode *r = new RegionNode(3); 4063 Node *phi = new PhiNode(r, TypeLong::LONG); 4064 4065 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4066 // Build the boolean node 4067 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4068 4069 // Branch either way. 4070 // NaN case is less traveled, which makes all the difference. 4071 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4072 Node *opt_isnan = _gvn.transform(ifisnan); 4073 assert( opt_isnan->is_If(), "Expect an IfNode"); 4074 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4075 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4076 4077 set_control(iftrue); 4078 4079 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4080 Node *slow_result = longcon(nan_bits); // return NaN 4081 phi->init_req(1, _gvn.transform( slow_result )); 4082 r->init_req(1, iftrue); 4083 4084 // Else fall through 4085 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4086 set_control(iffalse); 4087 4088 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4089 r->init_req(2, iffalse); 4090 4091 // Post merge 4092 set_control(_gvn.transform(r)); 4093 record_for_igvn(r); 4094 4095 C->set_has_split_ifs(true); // Has chance for split-if optimization 4096 result = phi; 4097 assert(result->bottom_type()->isa_long(), "must be"); 4098 break; 4099 } 4100 4101 case vmIntrinsics::_floatToIntBits: { 4102 // two paths (plus control) merge in a wood 4103 RegionNode *r = new RegionNode(3); 4104 Node *phi = new PhiNode(r, TypeInt::INT); 4105 4106 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 4107 // Build the boolean node 4108 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4109 4110 // Branch either way. 4111 // NaN case is less traveled, which makes all the difference. 4112 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4113 Node *opt_isnan = _gvn.transform(ifisnan); 4114 assert( opt_isnan->is_If(), "Expect an IfNode"); 4115 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4116 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4117 4118 set_control(iftrue); 4119 4120 static const jint nan_bits = 0x7fc00000; 4121 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 4122 phi->init_req(1, _gvn.transform( slow_result )); 4123 r->init_req(1, iftrue); 4124 4125 // Else fall through 4126 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4127 set_control(iffalse); 4128 4129 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 4130 r->init_req(2, iffalse); 4131 4132 // Post merge 4133 set_control(_gvn.transform(r)); 4134 record_for_igvn(r); 4135 4136 C->set_has_split_ifs(true); // Has chance for split-if optimization 4137 result = phi; 4138 assert(result->bottom_type()->isa_int(), "must be"); 4139 break; 4140 } 4141 4142 default: 4143 fatal_unexpected_iid(id); 4144 break; 4145 } 4146 set_result(_gvn.transform(result)); 4147 return true; 4148 } 4149 4150 //----------------------inline_unsafe_copyMemory------------------------- 4151 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 4152 bool LibraryCallKit::inline_unsafe_copyMemory() { 4153 if (callee()->is_static()) return false; // caller must have the capability! 4154 null_check_receiver(); // null-check receiver 4155 if (stopped()) return true; 4156 4157 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 4158 4159 Node* src_ptr = argument(1); // type: oop 4160 Node* src_off = ConvL2X(argument(2)); // type: long 4161 Node* dst_ptr = argument(4); // type: oop 4162 Node* dst_off = ConvL2X(argument(5)); // type: long 4163 Node* size = ConvL2X(argument(7)); // type: long 4164 4165 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 4166 "fieldOffset must be byte-scaled"); 4167 4168 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ); 4169 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE); 4170 4171 // Conservatively insert a memory barrier on all memory slices. 4172 // Do not let writes of the copy source or destination float below the copy. 4173 insert_mem_bar(Op_MemBarCPUOrder); 4174 4175 Node* thread = _gvn.transform(new ThreadLocalNode()); 4176 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 4177 BasicType doing_unsafe_access_bt = T_BYTE; 4178 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 4179 4180 // update volatile field 4181 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4182 4183 // Call it. Note that the length argument is not scaled. 4184 make_runtime_call(RC_LEAF|RC_NO_FP, 4185 OptoRuntime::fast_arraycopy_Type(), 4186 StubRoutines::unsafe_arraycopy(), 4187 "unsafe_arraycopy", 4188 TypeRawPtr::BOTTOM, 4189 src, dst, size XTOP); 4190 4191 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4192 4193 // Do not let reads of the copy destination float above the copy. 4194 insert_mem_bar(Op_MemBarCPUOrder); 4195 4196 return true; 4197 } 4198 4199 //------------------------clone_coping----------------------------------- 4200 // Helper function for inline_native_clone. 4201 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 4202 assert(obj_size != NULL, ""); 4203 Node* raw_obj = alloc_obj->in(1); 4204 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 4205 4206 AllocateNode* alloc = NULL; 4207 if (ReduceBulkZeroing) { 4208 // We will be completely responsible for initializing this object - 4209 // mark Initialize node as complete. 4210 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 4211 // The object was just allocated - there should be no any stores! 4212 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), ""); 4213 // Mark as complete_with_arraycopy so that on AllocateNode 4214 // expansion, we know this AllocateNode is initialized by an array 4215 // copy and a StoreStore barrier exists after the array copy. 4216 alloc->initialization()->set_complete_with_arraycopy(); 4217 } 4218 4219 Node* size = _gvn.transform(obj_size); 4220 access_clone(obj, alloc_obj, size, is_array); 4221 4222 // Do not let reads from the cloned object float above the arraycopy. 4223 if (alloc != NULL) { 4224 // Do not let stores that initialize this object be reordered with 4225 // a subsequent store that would make this object accessible by 4226 // other threads. 4227 // Record what AllocateNode this StoreStore protects so that 4228 // escape analysis can go from the MemBarStoreStoreNode to the 4229 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 4230 // based on the escape status of the AllocateNode. 4231 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 4232 } else { 4233 insert_mem_bar(Op_MemBarCPUOrder); 4234 } 4235 } 4236 4237 //------------------------inline_native_clone---------------------------- 4238 // protected native Object java.lang.Object.clone(); 4239 // 4240 // Here are the simple edge cases: 4241 // null receiver => normal trap 4242 // virtual and clone was overridden => slow path to out-of-line clone 4243 // not cloneable or finalizer => slow path to out-of-line Object.clone 4244 // 4245 // The general case has two steps, allocation and copying. 4246 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 4247 // 4248 // Copying also has two cases, oop arrays and everything else. 4249 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 4250 // Everything else uses the tight inline loop supplied by CopyArrayNode. 4251 // 4252 // These steps fold up nicely if and when the cloned object's klass 4253 // can be sharply typed as an object array, a type array, or an instance. 4254 // 4255 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 4256 PhiNode* result_val; 4257 4258 // Set the reexecute bit for the interpreter to reexecute 4259 // the bytecode that invokes Object.clone if deoptimization happens. 4260 { PreserveReexecuteState preexecs(this); 4261 jvms()->set_should_reexecute(true); 4262 4263 Node* obj = null_check_receiver(); 4264 if (stopped()) return true; 4265 4266 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4267 4268 // If we are going to clone an instance, we need its exact type to 4269 // know the number and types of fields to convert the clone to 4270 // loads/stores. Maybe a speculative type can help us. 4271 if (!obj_type->klass_is_exact() && 4272 obj_type->speculative_type() != NULL && 4273 obj_type->speculative_type()->is_instance_klass()) { 4274 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 4275 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 4276 !spec_ik->has_injected_fields()) { 4277 ciKlass* k = obj_type->klass(); 4278 if (!k->is_instance_klass() || 4279 k->as_instance_klass()->is_interface() || 4280 k->as_instance_klass()->has_subklass()) { 4281 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 4282 } 4283 } 4284 } 4285 4286 // Conservatively insert a memory barrier on all memory slices. 4287 // Do not let writes into the original float below the clone. 4288 insert_mem_bar(Op_MemBarCPUOrder); 4289 4290 // paths into result_reg: 4291 enum { 4292 _slow_path = 1, // out-of-line call to clone method (virtual or not) 4293 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 4294 _array_path, // plain array allocation, plus arrayof_long_arraycopy 4295 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 4296 PATH_LIMIT 4297 }; 4298 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4299 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4300 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 4301 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4302 record_for_igvn(result_reg); 4303 4304 Node* obj_klass = load_object_klass(obj); 4305 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL); 4306 if (array_ctl != NULL) { 4307 // It's an array. 4308 PreserveJVMState pjvms(this); 4309 set_control(array_ctl); 4310 Node* obj_length = load_array_length(obj); 4311 Node* obj_size = NULL; 4312 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push 4313 4314 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4315 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) { 4316 // If it is an oop array, it requires very special treatment, 4317 // because gc barriers are required when accessing the array. 4318 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL); 4319 if (is_obja != NULL) { 4320 PreserveJVMState pjvms2(this); 4321 set_control(is_obja); 4322 // Generate a direct call to the right arraycopy function(s). 4323 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL); 4324 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false); 4325 ac->set_clone_oop_array(); 4326 Node* n = _gvn.transform(ac); 4327 assert(n == ac, "cannot disappear"); 4328 ac->connect_outputs(this); 4329 4330 result_reg->init_req(_objArray_path, control()); 4331 result_val->init_req(_objArray_path, alloc_obj); 4332 result_i_o ->set_req(_objArray_path, i_o()); 4333 result_mem ->set_req(_objArray_path, reset_memory()); 4334 } 4335 } 4336 // Otherwise, there are no barriers to worry about. 4337 // (We can dispense with card marks if we know the allocation 4338 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 4339 // causes the non-eden paths to take compensating steps to 4340 // simulate a fresh allocation, so that no further 4341 // card marks are required in compiled code to initialize 4342 // the object.) 4343 4344 if (!stopped()) { 4345 copy_to_clone(obj, alloc_obj, obj_size, true); 4346 4347 // Present the results of the copy. 4348 result_reg->init_req(_array_path, control()); 4349 result_val->init_req(_array_path, alloc_obj); 4350 result_i_o ->set_req(_array_path, i_o()); 4351 result_mem ->set_req(_array_path, reset_memory()); 4352 } 4353 } 4354 4355 // We only go to the instance fast case code if we pass a number of guards. 4356 // The paths which do not pass are accumulated in the slow_region. 4357 RegionNode* slow_region = new RegionNode(1); 4358 record_for_igvn(slow_region); 4359 if (!stopped()) { 4360 // It's an instance (we did array above). Make the slow-path tests. 4361 // If this is a virtual call, we generate a funny guard. We grab 4362 // the vtable entry corresponding to clone() from the target object. 4363 // If the target method which we are calling happens to be the 4364 // Object clone() method, we pass the guard. We do not need this 4365 // guard for non-virtual calls; the caller is known to be the native 4366 // Object clone(). 4367 if (is_virtual) { 4368 generate_virtual_guard(obj_klass, slow_region); 4369 } 4370 4371 // The object must be easily cloneable and must not have a finalizer. 4372 // Both of these conditions may be checked in a single test. 4373 // We could optimize the test further, but we don't care. 4374 generate_access_flags_guard(obj_klass, 4375 // Test both conditions: 4376 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 4377 // Must be cloneable but not finalizer: 4378 JVM_ACC_IS_CLONEABLE_FAST, 4379 slow_region); 4380 } 4381 4382 if (!stopped()) { 4383 // It's an instance, and it passed the slow-path tests. 4384 PreserveJVMState pjvms(this); 4385 Node* obj_size = NULL; 4386 // Need to deoptimize on exception from allocation since Object.clone intrinsic 4387 // is reexecuted if deoptimization occurs and there could be problems when merging 4388 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 4389 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 4390 4391 copy_to_clone(obj, alloc_obj, obj_size, false); 4392 4393 // Present the results of the slow call. 4394 result_reg->init_req(_instance_path, control()); 4395 result_val->init_req(_instance_path, alloc_obj); 4396 result_i_o ->set_req(_instance_path, i_o()); 4397 result_mem ->set_req(_instance_path, reset_memory()); 4398 } 4399 4400 // Generate code for the slow case. We make a call to clone(). 4401 set_control(_gvn.transform(slow_region)); 4402 if (!stopped()) { 4403 PreserveJVMState pjvms(this); 4404 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual); 4405 // We need to deoptimize on exception (see comment above) 4406 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 4407 // this->control() comes from set_results_for_java_call 4408 result_reg->init_req(_slow_path, control()); 4409 result_val->init_req(_slow_path, slow_result); 4410 result_i_o ->set_req(_slow_path, i_o()); 4411 result_mem ->set_req(_slow_path, reset_memory()); 4412 } 4413 4414 // Return the combined state. 4415 set_control( _gvn.transform(result_reg)); 4416 set_i_o( _gvn.transform(result_i_o)); 4417 set_all_memory( _gvn.transform(result_mem)); 4418 } // original reexecute is set back here 4419 4420 set_result(_gvn.transform(result_val)); 4421 return true; 4422 } 4423 4424 // If we have a tightly coupled allocation, the arraycopy may take care 4425 // of the array initialization. If one of the guards we insert between 4426 // the allocation and the arraycopy causes a deoptimization, an 4427 // unitialized array will escape the compiled method. To prevent that 4428 // we set the JVM state for uncommon traps between the allocation and 4429 // the arraycopy to the state before the allocation so, in case of 4430 // deoptimization, we'll reexecute the allocation and the 4431 // initialization. 4432 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 4433 if (alloc != NULL) { 4434 ciMethod* trap_method = alloc->jvms()->method(); 4435 int trap_bci = alloc->jvms()->bci(); 4436 4437 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4438 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 4439 // Make sure there's no store between the allocation and the 4440 // arraycopy otherwise visible side effects could be rexecuted 4441 // in case of deoptimization and cause incorrect execution. 4442 bool no_interfering_store = true; 4443 Node* mem = alloc->in(TypeFunc::Memory); 4444 if (mem->is_MergeMem()) { 4445 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 4446 Node* n = mms.memory(); 4447 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4448 assert(n->is_Store(), "what else?"); 4449 no_interfering_store = false; 4450 break; 4451 } 4452 } 4453 } else { 4454 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 4455 Node* n = mms.memory(); 4456 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4457 assert(n->is_Store(), "what else?"); 4458 no_interfering_store = false; 4459 break; 4460 } 4461 } 4462 } 4463 4464 if (no_interfering_store) { 4465 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 4466 uint size = alloc->req(); 4467 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 4468 old_jvms->set_map(sfpt); 4469 for (uint i = 0; i < size; i++) { 4470 sfpt->init_req(i, alloc->in(i)); 4471 } 4472 // re-push array length for deoptimization 4473 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 4474 old_jvms->set_sp(old_jvms->sp()+1); 4475 old_jvms->set_monoff(old_jvms->monoff()+1); 4476 old_jvms->set_scloff(old_jvms->scloff()+1); 4477 old_jvms->set_endoff(old_jvms->endoff()+1); 4478 old_jvms->set_should_reexecute(true); 4479 4480 sfpt->set_i_o(map()->i_o()); 4481 sfpt->set_memory(map()->memory()); 4482 sfpt->set_control(map()->control()); 4483 4484 JVMState* saved_jvms = jvms(); 4485 saved_reexecute_sp = _reexecute_sp; 4486 4487 set_jvms(sfpt->jvms()); 4488 _reexecute_sp = jvms()->sp(); 4489 4490 return saved_jvms; 4491 } 4492 } 4493 } 4494 return NULL; 4495 } 4496 4497 // In case of a deoptimization, we restart execution at the 4498 // allocation, allocating a new array. We would leave an uninitialized 4499 // array in the heap that GCs wouldn't expect. Move the allocation 4500 // after the traps so we don't allocate the array if we 4501 // deoptimize. This is possible because tightly_coupled_allocation() 4502 // guarantees there's no observer of the allocated array at this point 4503 // and the control flow is simple enough. 4504 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, 4505 int saved_reexecute_sp, uint new_idx) { 4506 if (saved_jvms != NULL && !stopped()) { 4507 assert(alloc != NULL, "only with a tightly coupled allocation"); 4508 // restore JVM state to the state at the arraycopy 4509 saved_jvms->map()->set_control(map()->control()); 4510 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?"); 4511 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?"); 4512 // If we've improved the types of some nodes (null check) while 4513 // emitting the guards, propagate them to the current state 4514 map()->replaced_nodes().apply(saved_jvms->map(), new_idx); 4515 set_jvms(saved_jvms); 4516 _reexecute_sp = saved_reexecute_sp; 4517 4518 // Remove the allocation from above the guards 4519 CallProjections callprojs; 4520 alloc->extract_projections(&callprojs, true); 4521 InitializeNode* init = alloc->initialization(); 4522 Node* alloc_mem = alloc->in(TypeFunc::Memory); 4523 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 4524 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 4525 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 4526 4527 // move the allocation here (after the guards) 4528 _gvn.hash_delete(alloc); 4529 alloc->set_req(TypeFunc::Control, control()); 4530 alloc->set_req(TypeFunc::I_O, i_o()); 4531 Node *mem = reset_memory(); 4532 set_all_memory(mem); 4533 alloc->set_req(TypeFunc::Memory, mem); 4534 set_control(init->proj_out_or_null(TypeFunc::Control)); 4535 set_i_o(callprojs.fallthrough_ioproj); 4536 4537 // Update memory as done in GraphKit::set_output_for_allocation() 4538 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 4539 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 4540 if (ary_type->isa_aryptr() && length_type != NULL) { 4541 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 4542 } 4543 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 4544 int elemidx = C->get_alias_index(telemref); 4545 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 4546 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 4547 4548 Node* allocx = _gvn.transform(alloc); 4549 assert(allocx == alloc, "where has the allocation gone?"); 4550 assert(dest->is_CheckCastPP(), "not an allocation result?"); 4551 4552 _gvn.hash_delete(dest); 4553 dest->set_req(0, control()); 4554 Node* destx = _gvn.transform(dest); 4555 assert(destx == dest, "where has the allocation result gone?"); 4556 } 4557 } 4558 4559 4560 //------------------------------inline_arraycopy----------------------- 4561 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 4562 // Object dest, int destPos, 4563 // int length); 4564 bool LibraryCallKit::inline_arraycopy() { 4565 // Get the arguments. 4566 Node* src = argument(0); // type: oop 4567 Node* src_offset = argument(1); // type: int 4568 Node* dest = argument(2); // type: oop 4569 Node* dest_offset = argument(3); // type: int 4570 Node* length = argument(4); // type: int 4571 4572 uint new_idx = C->unique(); 4573 4574 // Check for allocation before we add nodes that would confuse 4575 // tightly_coupled_allocation() 4576 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL); 4577 4578 int saved_reexecute_sp = -1; 4579 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 4580 // See arraycopy_restore_alloc_state() comment 4581 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards 4582 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation 4583 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards 4584 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL); 4585 4586 // The following tests must be performed 4587 // (1) src and dest are arrays. 4588 // (2) src and dest arrays must have elements of the same BasicType 4589 // (3) src and dest must not be null. 4590 // (4) src_offset must not be negative. 4591 // (5) dest_offset must not be negative. 4592 // (6) length must not be negative. 4593 // (7) src_offset + length must not exceed length of src. 4594 // (8) dest_offset + length must not exceed length of dest. 4595 // (9) each element of an oop array must be assignable 4596 4597 // (3) src and dest must not be null. 4598 // always do this here because we need the JVM state for uncommon traps 4599 Node* null_ctl = top(); 4600 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 4601 assert(null_ctl->is_top(), "no null control here"); 4602 dest = null_check(dest, T_ARRAY); 4603 4604 if (!can_emit_guards) { 4605 // if saved_jvms == NULL and alloc != NULL, we don't emit any 4606 // guards but the arraycopy node could still take advantage of a 4607 // tightly allocated allocation. tightly_coupled_allocation() is 4608 // called again to make sure it takes the null check above into 4609 // account: the null check is mandatory and if it caused an 4610 // uncommon trap to be emitted then the allocation can't be 4611 // considered tightly coupled in this context. 4612 alloc = tightly_coupled_allocation(dest, NULL); 4613 } 4614 4615 bool validated = false; 4616 4617 const Type* src_type = _gvn.type(src); 4618 const Type* dest_type = _gvn.type(dest); 4619 const TypeAryPtr* top_src = src_type->isa_aryptr(); 4620 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 4621 4622 // Do we have the type of src? 4623 bool has_src = (top_src != NULL && top_src->klass() != NULL); 4624 // Do we have the type of dest? 4625 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL); 4626 // Is the type for src from speculation? 4627 bool src_spec = false; 4628 // Is the type for dest from speculation? 4629 bool dest_spec = false; 4630 4631 if ((!has_src || !has_dest) && can_emit_guards) { 4632 // We don't have sufficient type information, let's see if 4633 // speculative types can help. We need to have types for both src 4634 // and dest so that it pays off. 4635 4636 // Do we already have or could we have type information for src 4637 bool could_have_src = has_src; 4638 // Do we already have or could we have type information for dest 4639 bool could_have_dest = has_dest; 4640 4641 ciKlass* src_k = NULL; 4642 if (!has_src) { 4643 src_k = src_type->speculative_type_not_null(); 4644 if (src_k != NULL && src_k->is_array_klass()) { 4645 could_have_src = true; 4646 } 4647 } 4648 4649 ciKlass* dest_k = NULL; 4650 if (!has_dest) { 4651 dest_k = dest_type->speculative_type_not_null(); 4652 if (dest_k != NULL && dest_k->is_array_klass()) { 4653 could_have_dest = true; 4654 } 4655 } 4656 4657 if (could_have_src && could_have_dest) { 4658 // This is going to pay off so emit the required guards 4659 if (!has_src) { 4660 src = maybe_cast_profiled_obj(src, src_k, true); 4661 src_type = _gvn.type(src); 4662 top_src = src_type->isa_aryptr(); 4663 has_src = (top_src != NULL && top_src->klass() != NULL); 4664 src_spec = true; 4665 } 4666 if (!has_dest) { 4667 dest = maybe_cast_profiled_obj(dest, dest_k, true); 4668 dest_type = _gvn.type(dest); 4669 top_dest = dest_type->isa_aryptr(); 4670 has_dest = (top_dest != NULL && top_dest->klass() != NULL); 4671 dest_spec = true; 4672 } 4673 } 4674 } 4675 4676 if (has_src && has_dest && can_emit_guards) { 4677 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type(); 4678 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type(); 4679 if (is_reference_type(src_elem)) src_elem = T_OBJECT; 4680 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT; 4681 4682 if (src_elem == dest_elem && src_elem == T_OBJECT) { 4683 // If both arrays are object arrays then having the exact types 4684 // for both will remove the need for a subtype check at runtime 4685 // before the call and may make it possible to pick a faster copy 4686 // routine (without a subtype check on every element) 4687 // Do we have the exact type of src? 4688 bool could_have_src = src_spec; 4689 // Do we have the exact type of dest? 4690 bool could_have_dest = dest_spec; 4691 ciKlass* src_k = top_src->klass(); 4692 ciKlass* dest_k = top_dest->klass(); 4693 if (!src_spec) { 4694 src_k = src_type->speculative_type_not_null(); 4695 if (src_k != NULL && src_k->is_array_klass()) { 4696 could_have_src = true; 4697 } 4698 } 4699 if (!dest_spec) { 4700 dest_k = dest_type->speculative_type_not_null(); 4701 if (dest_k != NULL && dest_k->is_array_klass()) { 4702 could_have_dest = true; 4703 } 4704 } 4705 if (could_have_src && could_have_dest) { 4706 // If we can have both exact types, emit the missing guards 4707 if (could_have_src && !src_spec) { 4708 src = maybe_cast_profiled_obj(src, src_k, true); 4709 } 4710 if (could_have_dest && !dest_spec) { 4711 dest = maybe_cast_profiled_obj(dest, dest_k, true); 4712 } 4713 } 4714 } 4715 } 4716 4717 ciMethod* trap_method = method(); 4718 int trap_bci = bci(); 4719 if (saved_jvms != NULL) { 4720 trap_method = alloc->jvms()->method(); 4721 trap_bci = alloc->jvms()->bci(); 4722 } 4723 4724 bool negative_length_guard_generated = false; 4725 4726 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4727 can_emit_guards && 4728 !src->is_top() && !dest->is_top()) { 4729 // validate arguments: enables transformation the ArrayCopyNode 4730 validated = true; 4731 4732 RegionNode* slow_region = new RegionNode(1); 4733 record_for_igvn(slow_region); 4734 4735 // (1) src and dest are arrays. 4736 generate_non_array_guard(load_object_klass(src), slow_region); 4737 generate_non_array_guard(load_object_klass(dest), slow_region); 4738 4739 // (2) src and dest arrays must have elements of the same BasicType 4740 // done at macro expansion or at Ideal transformation time 4741 4742 // (4) src_offset must not be negative. 4743 generate_negative_guard(src_offset, slow_region); 4744 4745 // (5) dest_offset must not be negative. 4746 generate_negative_guard(dest_offset, slow_region); 4747 4748 // (7) src_offset + length must not exceed length of src. 4749 generate_limit_guard(src_offset, length, 4750 load_array_length(src), 4751 slow_region); 4752 4753 // (8) dest_offset + length must not exceed length of dest. 4754 generate_limit_guard(dest_offset, length, 4755 load_array_length(dest), 4756 slow_region); 4757 4758 // (6) length must not be negative. 4759 // This is also checked in generate_arraycopy() during macro expansion, but 4760 // we also have to check it here for the case where the ArrayCopyNode will 4761 // be eliminated by Escape Analysis. 4762 if (EliminateAllocations) { 4763 generate_negative_guard(length, slow_region); 4764 negative_length_guard_generated = true; 4765 } 4766 4767 // (9) each element of an oop array must be assignable 4768 Node* dest_klass = load_object_klass(dest); 4769 if (src != dest) { 4770 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass); 4771 4772 if (not_subtype_ctrl != top()) { 4773 PreserveJVMState pjvms(this); 4774 set_control(not_subtype_ctrl); 4775 uncommon_trap(Deoptimization::Reason_intrinsic, 4776 Deoptimization::Action_make_not_entrant); 4777 assert(stopped(), "Should be stopped"); 4778 } 4779 } 4780 { 4781 PreserveJVMState pjvms(this); 4782 set_control(_gvn.transform(slow_region)); 4783 uncommon_trap(Deoptimization::Reason_intrinsic, 4784 Deoptimization::Action_make_not_entrant); 4785 assert(stopped(), "Should be stopped"); 4786 } 4787 4788 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 4789 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass()); 4790 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 4791 } 4792 4793 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx); 4794 4795 if (stopped()) { 4796 return true; 4797 } 4798 4799 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated, 4800 // Create LoadRange and LoadKlass nodes for use during macro expansion here 4801 // so the compiler has a chance to eliminate them: during macro expansion, 4802 // we have to set their control (CastPP nodes are eliminated). 4803 load_object_klass(src), load_object_klass(dest), 4804 load_array_length(src), load_array_length(dest)); 4805 4806 ac->set_arraycopy(validated); 4807 4808 Node* n = _gvn.transform(ac); 4809 if (n == ac) { 4810 ac->connect_outputs(this); 4811 } else { 4812 assert(validated, "shouldn't transform if all arguments not validated"); 4813 set_all_memory(n); 4814 } 4815 clear_upper_avx(); 4816 4817 4818 return true; 4819 } 4820 4821 4822 // Helper function which determines if an arraycopy immediately follows 4823 // an allocation, with no intervening tests or other escapes for the object. 4824 AllocateArrayNode* 4825 LibraryCallKit::tightly_coupled_allocation(Node* ptr, 4826 RegionNode* slow_region) { 4827 if (stopped()) return NULL; // no fast path 4828 if (C->AliasLevel() == 0) return NULL; // no MergeMems around 4829 4830 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn); 4831 if (alloc == NULL) return NULL; 4832 4833 Node* rawmem = memory(Compile::AliasIdxRaw); 4834 // Is the allocation's memory state untouched? 4835 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 4836 // Bail out if there have been raw-memory effects since the allocation. 4837 // (Example: There might have been a call or safepoint.) 4838 return NULL; 4839 } 4840 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 4841 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 4842 return NULL; 4843 } 4844 4845 // There must be no unexpected observers of this allocation. 4846 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 4847 Node* obs = ptr->fast_out(i); 4848 if (obs != this->map()) { 4849 return NULL; 4850 } 4851 } 4852 4853 // This arraycopy must unconditionally follow the allocation of the ptr. 4854 Node* alloc_ctl = ptr->in(0); 4855 Node* ctl = control(); 4856 while (ctl != alloc_ctl) { 4857 // There may be guards which feed into the slow_region. 4858 // Any other control flow means that we might not get a chance 4859 // to finish initializing the allocated object. 4860 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) { 4861 IfNode* iff = ctl->in(0)->as_If(); 4862 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con); 4863 assert(not_ctl != NULL && not_ctl != ctl, "found alternate"); 4864 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) { 4865 ctl = iff->in(0); // This test feeds the known slow_region. 4866 continue; 4867 } 4868 // One more try: Various low-level checks bottom out in 4869 // uncommon traps. If the debug-info of the trap omits 4870 // any reference to the allocation, as we've already 4871 // observed, then there can be no objection to the trap. 4872 bool found_trap = false; 4873 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) { 4874 Node* obs = not_ctl->fast_out(j); 4875 if (obs->in(0) == not_ctl && obs->is_Call() && 4876 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 4877 found_trap = true; break; 4878 } 4879 } 4880 if (found_trap) { 4881 ctl = iff->in(0); // This test feeds a harmless uncommon trap. 4882 continue; 4883 } 4884 } 4885 return NULL; 4886 } 4887 4888 // If we get this far, we have an allocation which immediately 4889 // precedes the arraycopy, and we can take over zeroing the new object. 4890 // The arraycopy will finish the initialization, and provide 4891 // a new control state to which we will anchor the destination pointer. 4892 4893 return alloc; 4894 } 4895 4896 //-------------inline_encodeISOArray----------------------------------- 4897 // encode char[] to byte[] in ISO_8859_1 4898 bool LibraryCallKit::inline_encodeISOArray() { 4899 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 4900 // no receiver since it is static method 4901 Node *src = argument(0); 4902 Node *src_offset = argument(1); 4903 Node *dst = argument(2); 4904 Node *dst_offset = argument(3); 4905 Node *length = argument(4); 4906 4907 src = must_be_not_null(src, true); 4908 dst = must_be_not_null(dst, true); 4909 4910 const Type* src_type = src->Value(&_gvn); 4911 const Type* dst_type = dst->Value(&_gvn); 4912 const TypeAryPtr* top_src = src_type->isa_aryptr(); 4913 const TypeAryPtr* top_dest = dst_type->isa_aryptr(); 4914 if (top_src == NULL || top_src->klass() == NULL || 4915 top_dest == NULL || top_dest->klass() == NULL) { 4916 // failed array check 4917 return false; 4918 } 4919 4920 // Figure out the size and type of the elements we will be copying. 4921 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4922 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4923 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 4924 return false; 4925 } 4926 4927 Node* src_start = array_element_address(src, src_offset, T_CHAR); 4928 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 4929 // 'src_start' points to src array + scaled offset 4930 // 'dst_start' points to dst array + scaled offset 4931 4932 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 4933 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length); 4934 enc = _gvn.transform(enc); 4935 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 4936 set_memory(res_mem, mtype); 4937 set_result(enc); 4938 clear_upper_avx(); 4939 4940 return true; 4941 } 4942 4943 //-------------inline_multiplyToLen----------------------------------- 4944 bool LibraryCallKit::inline_multiplyToLen() { 4945 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 4946 4947 address stubAddr = StubRoutines::multiplyToLen(); 4948 if (stubAddr == NULL) { 4949 return false; // Intrinsic's stub is not implemented on this platform 4950 } 4951 const char* stubName = "multiplyToLen"; 4952 4953 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 4954 4955 // no receiver because it is a static method 4956 Node* x = argument(0); 4957 Node* xlen = argument(1); 4958 Node* y = argument(2); 4959 Node* ylen = argument(3); 4960 Node* z = argument(4); 4961 4962 x = must_be_not_null(x, true); 4963 y = must_be_not_null(y, true); 4964 4965 const Type* x_type = x->Value(&_gvn); 4966 const Type* y_type = y->Value(&_gvn); 4967 const TypeAryPtr* top_x = x_type->isa_aryptr(); 4968 const TypeAryPtr* top_y = y_type->isa_aryptr(); 4969 if (top_x == NULL || top_x->klass() == NULL || 4970 top_y == NULL || top_y->klass() == NULL) { 4971 // failed array check 4972 return false; 4973 } 4974 4975 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4976 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4977 if (x_elem != T_INT || y_elem != T_INT) { 4978 return false; 4979 } 4980 4981 // Set the original stack and the reexecute bit for the interpreter to reexecute 4982 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 4983 // on the return from z array allocation in runtime. 4984 { PreserveReexecuteState preexecs(this); 4985 jvms()->set_should_reexecute(true); 4986 4987 Node* x_start = array_element_address(x, intcon(0), x_elem); 4988 Node* y_start = array_element_address(y, intcon(0), y_elem); 4989 // 'x_start' points to x array + scaled xlen 4990 // 'y_start' points to y array + scaled ylen 4991 4992 // Allocate the result array 4993 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 4994 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 4995 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 4996 4997 IdealKit ideal(this); 4998 4999 #define __ ideal. 5000 Node* one = __ ConI(1); 5001 Node* zero = __ ConI(0); 5002 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 5003 __ set(need_alloc, zero); 5004 __ set(z_alloc, z); 5005 __ if_then(z, BoolTest::eq, null()); { 5006 __ increment (need_alloc, one); 5007 } __ else_(); { 5008 // Update graphKit memory and control from IdealKit. 5009 sync_kit(ideal); 5010 Node *cast = new CastPPNode(z, TypePtr::NOTNULL); 5011 cast->init_req(0, control()); 5012 _gvn.set_type(cast, cast->bottom_type()); 5013 C->record_for_igvn(cast); 5014 5015 Node* zlen_arg = load_array_length(cast); 5016 // Update IdealKit memory and control from graphKit. 5017 __ sync_kit(this); 5018 __ if_then(zlen_arg, BoolTest::lt, zlen); { 5019 __ increment (need_alloc, one); 5020 } __ end_if(); 5021 } __ end_if(); 5022 5023 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 5024 // Update graphKit memory and control from IdealKit. 5025 sync_kit(ideal); 5026 Node * narr = new_array(klass_node, zlen, 1); 5027 // Update IdealKit memory and control from graphKit. 5028 __ sync_kit(this); 5029 __ set(z_alloc, narr); 5030 } __ end_if(); 5031 5032 sync_kit(ideal); 5033 z = __ value(z_alloc); 5034 // Can't use TypeAryPtr::INTS which uses Bottom offset. 5035 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 5036 // Final sync IdealKit and GraphKit. 5037 final_sync(ideal); 5038 #undef __ 5039 5040 Node* z_start = array_element_address(z, intcon(0), T_INT); 5041 5042 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5043 OptoRuntime::multiplyToLen_Type(), 5044 stubAddr, stubName, TypePtr::BOTTOM, 5045 x_start, xlen, y_start, ylen, z_start, zlen); 5046 } // original reexecute is set back here 5047 5048 C->set_has_split_ifs(true); // Has chance for split-if optimization 5049 set_result(z); 5050 return true; 5051 } 5052 5053 //-------------inline_squareToLen------------------------------------ 5054 bool LibraryCallKit::inline_squareToLen() { 5055 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 5056 5057 address stubAddr = StubRoutines::squareToLen(); 5058 if (stubAddr == NULL) { 5059 return false; // Intrinsic's stub is not implemented on this platform 5060 } 5061 const char* stubName = "squareToLen"; 5062 5063 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 5064 5065 Node* x = argument(0); 5066 Node* len = argument(1); 5067 Node* z = argument(2); 5068 Node* zlen = argument(3); 5069 5070 x = must_be_not_null(x, true); 5071 z = must_be_not_null(z, true); 5072 5073 const Type* x_type = x->Value(&_gvn); 5074 const Type* z_type = z->Value(&_gvn); 5075 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5076 const TypeAryPtr* top_z = z_type->isa_aryptr(); 5077 if (top_x == NULL || top_x->klass() == NULL || 5078 top_z == NULL || top_z->klass() == NULL) { 5079 // failed array check 5080 return false; 5081 } 5082 5083 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5084 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5085 if (x_elem != T_INT || z_elem != T_INT) { 5086 return false; 5087 } 5088 5089 5090 Node* x_start = array_element_address(x, intcon(0), x_elem); 5091 Node* z_start = array_element_address(z, intcon(0), z_elem); 5092 5093 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5094 OptoRuntime::squareToLen_Type(), 5095 stubAddr, stubName, TypePtr::BOTTOM, 5096 x_start, len, z_start, zlen); 5097 5098 set_result(z); 5099 return true; 5100 } 5101 5102 //-------------inline_mulAdd------------------------------------------ 5103 bool LibraryCallKit::inline_mulAdd() { 5104 assert(UseMulAddIntrinsic, "not implemented on this platform"); 5105 5106 address stubAddr = StubRoutines::mulAdd(); 5107 if (stubAddr == NULL) { 5108 return false; // Intrinsic's stub is not implemented on this platform 5109 } 5110 const char* stubName = "mulAdd"; 5111 5112 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 5113 5114 Node* out = argument(0); 5115 Node* in = argument(1); 5116 Node* offset = argument(2); 5117 Node* len = argument(3); 5118 Node* k = argument(4); 5119 5120 out = must_be_not_null(out, true); 5121 5122 const Type* out_type = out->Value(&_gvn); 5123 const Type* in_type = in->Value(&_gvn); 5124 const TypeAryPtr* top_out = out_type->isa_aryptr(); 5125 const TypeAryPtr* top_in = in_type->isa_aryptr(); 5126 if (top_out == NULL || top_out->klass() == NULL || 5127 top_in == NULL || top_in->klass() == NULL) { 5128 // failed array check 5129 return false; 5130 } 5131 5132 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5133 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5134 if (out_elem != T_INT || in_elem != T_INT) { 5135 return false; 5136 } 5137 5138 Node* outlen = load_array_length(out); 5139 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 5140 Node* out_start = array_element_address(out, intcon(0), out_elem); 5141 Node* in_start = array_element_address(in, intcon(0), in_elem); 5142 5143 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5144 OptoRuntime::mulAdd_Type(), 5145 stubAddr, stubName, TypePtr::BOTTOM, 5146 out_start,in_start, new_offset, len, k); 5147 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5148 set_result(result); 5149 return true; 5150 } 5151 5152 //-------------inline_montgomeryMultiply----------------------------------- 5153 bool LibraryCallKit::inline_montgomeryMultiply() { 5154 address stubAddr = StubRoutines::montgomeryMultiply(); 5155 if (stubAddr == NULL) { 5156 return false; // Intrinsic's stub is not implemented on this platform 5157 } 5158 5159 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 5160 const char* stubName = "montgomery_multiply"; 5161 5162 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 5163 5164 Node* a = argument(0); 5165 Node* b = argument(1); 5166 Node* n = argument(2); 5167 Node* len = argument(3); 5168 Node* inv = argument(4); 5169 Node* m = argument(6); 5170 5171 const Type* a_type = a->Value(&_gvn); 5172 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5173 const Type* b_type = b->Value(&_gvn); 5174 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5175 const Type* n_type = a->Value(&_gvn); 5176 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5177 const Type* m_type = a->Value(&_gvn); 5178 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5179 if (top_a == NULL || top_a->klass() == NULL || 5180 top_b == NULL || top_b->klass() == NULL || 5181 top_n == NULL || top_n->klass() == NULL || 5182 top_m == NULL || top_m->klass() == NULL) { 5183 // failed array check 5184 return false; 5185 } 5186 5187 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5188 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5189 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5190 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5191 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5192 return false; 5193 } 5194 5195 // Make the call 5196 { 5197 Node* a_start = array_element_address(a, intcon(0), a_elem); 5198 Node* b_start = array_element_address(b, intcon(0), b_elem); 5199 Node* n_start = array_element_address(n, intcon(0), n_elem); 5200 Node* m_start = array_element_address(m, intcon(0), m_elem); 5201 5202 Node* call = make_runtime_call(RC_LEAF, 5203 OptoRuntime::montgomeryMultiply_Type(), 5204 stubAddr, stubName, TypePtr::BOTTOM, 5205 a_start, b_start, n_start, len, inv, top(), 5206 m_start); 5207 set_result(m); 5208 } 5209 5210 return true; 5211 } 5212 5213 bool LibraryCallKit::inline_montgomerySquare() { 5214 address stubAddr = StubRoutines::montgomerySquare(); 5215 if (stubAddr == NULL) { 5216 return false; // Intrinsic's stub is not implemented on this platform 5217 } 5218 5219 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 5220 const char* stubName = "montgomery_square"; 5221 5222 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 5223 5224 Node* a = argument(0); 5225 Node* n = argument(1); 5226 Node* len = argument(2); 5227 Node* inv = argument(3); 5228 Node* m = argument(5); 5229 5230 const Type* a_type = a->Value(&_gvn); 5231 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5232 const Type* n_type = a->Value(&_gvn); 5233 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5234 const Type* m_type = a->Value(&_gvn); 5235 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5236 if (top_a == NULL || top_a->klass() == NULL || 5237 top_n == NULL || top_n->klass() == NULL || 5238 top_m == NULL || top_m->klass() == NULL) { 5239 // failed array check 5240 return false; 5241 } 5242 5243 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5244 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5245 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5246 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5247 return false; 5248 } 5249 5250 // Make the call 5251 { 5252 Node* a_start = array_element_address(a, intcon(0), a_elem); 5253 Node* n_start = array_element_address(n, intcon(0), n_elem); 5254 Node* m_start = array_element_address(m, intcon(0), m_elem); 5255 5256 Node* call = make_runtime_call(RC_LEAF, 5257 OptoRuntime::montgomerySquare_Type(), 5258 stubAddr, stubName, TypePtr::BOTTOM, 5259 a_start, n_start, len, inv, top(), 5260 m_start); 5261 set_result(m); 5262 } 5263 5264 return true; 5265 } 5266 5267 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) { 5268 address stubAddr = NULL; 5269 const char* stubName = NULL; 5270 5271 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift(); 5272 if (stubAddr == NULL) { 5273 return false; // Intrinsic's stub is not implemented on this platform 5274 } 5275 5276 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker"; 5277 5278 assert(callee()->signature()->size() == 5, "expected 5 arguments"); 5279 5280 Node* newArr = argument(0); 5281 Node* oldArr = argument(1); 5282 Node* newIdx = argument(2); 5283 Node* shiftCount = argument(3); 5284 Node* numIter = argument(4); 5285 5286 const Type* newArr_type = newArr->Value(&_gvn); 5287 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr(); 5288 const Type* oldArr_type = oldArr->Value(&_gvn); 5289 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr(); 5290 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL 5291 || top_oldArr->klass() == NULL) { 5292 return false; 5293 } 5294 5295 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5296 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5297 if (newArr_elem != T_INT || oldArr_elem != T_INT) { 5298 return false; 5299 } 5300 5301 // Make the call 5302 { 5303 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem); 5304 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem); 5305 5306 Node* call = make_runtime_call(RC_LEAF, 5307 OptoRuntime::bigIntegerShift_Type(), 5308 stubAddr, 5309 stubName, 5310 TypePtr::BOTTOM, 5311 newArr_start, 5312 oldArr_start, 5313 newIdx, 5314 shiftCount, 5315 numIter); 5316 } 5317 5318 return true; 5319 } 5320 5321 //-------------inline_vectorizedMismatch------------------------------ 5322 bool LibraryCallKit::inline_vectorizedMismatch() { 5323 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform"); 5324 5325 address stubAddr = StubRoutines::vectorizedMismatch(); 5326 if (stubAddr == NULL) { 5327 return false; // Intrinsic's stub is not implemented on this platform 5328 } 5329 const char* stubName = "vectorizedMismatch"; 5330 int size_l = callee()->signature()->size(); 5331 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 5332 5333 Node* obja = argument(0); 5334 Node* aoffset = argument(1); 5335 Node* objb = argument(3); 5336 Node* boffset = argument(4); 5337 Node* length = argument(6); 5338 Node* scale = argument(7); 5339 5340 const Type* a_type = obja->Value(&_gvn); 5341 const Type* b_type = objb->Value(&_gvn); 5342 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5343 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5344 if (top_a == NULL || top_a->klass() == NULL || 5345 top_b == NULL || top_b->klass() == NULL) { 5346 // failed array check 5347 return false; 5348 } 5349 5350 Node* call; 5351 jvms()->set_should_reexecute(true); 5352 5353 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ); 5354 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ); 5355 5356 call = make_runtime_call(RC_LEAF, 5357 OptoRuntime::vectorizedMismatch_Type(), 5358 stubAddr, stubName, TypePtr::BOTTOM, 5359 obja_adr, objb_adr, length, scale); 5360 5361 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5362 set_result(result); 5363 return true; 5364 } 5365 5366 /** 5367 * Calculate CRC32 for byte. 5368 * int java.util.zip.CRC32.update(int crc, int b) 5369 */ 5370 bool LibraryCallKit::inline_updateCRC32() { 5371 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5372 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 5373 // no receiver since it is static method 5374 Node* crc = argument(0); // type: int 5375 Node* b = argument(1); // type: int 5376 5377 /* 5378 * int c = ~ crc; 5379 * b = timesXtoThe32[(b ^ c) & 0xFF]; 5380 * b = b ^ (c >>> 8); 5381 * crc = ~b; 5382 */ 5383 5384 Node* M1 = intcon(-1); 5385 crc = _gvn.transform(new XorINode(crc, M1)); 5386 Node* result = _gvn.transform(new XorINode(crc, b)); 5387 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 5388 5389 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 5390 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 5391 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 5392 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 5393 5394 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 5395 result = _gvn.transform(new XorINode(crc, result)); 5396 result = _gvn.transform(new XorINode(result, M1)); 5397 set_result(result); 5398 return true; 5399 } 5400 5401 /** 5402 * Calculate CRC32 for byte[] array. 5403 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 5404 */ 5405 bool LibraryCallKit::inline_updateBytesCRC32() { 5406 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5407 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5408 // no receiver since it is static method 5409 Node* crc = argument(0); // type: int 5410 Node* src = argument(1); // type: oop 5411 Node* offset = argument(2); // type: int 5412 Node* length = argument(3); // type: int 5413 5414 const Type* src_type = src->Value(&_gvn); 5415 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5416 if (top_src == NULL || top_src->klass() == NULL) { 5417 // failed array check 5418 return false; 5419 } 5420 5421 // Figure out the size and type of the elements we will be copying. 5422 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5423 if (src_elem != T_BYTE) { 5424 return false; 5425 } 5426 5427 // 'src_start' points to src array + scaled offset 5428 src = must_be_not_null(src, true); 5429 Node* src_start = array_element_address(src, offset, src_elem); 5430 5431 // We assume that range check is done by caller. 5432 // TODO: generate range check (offset+length < src.length) in debug VM. 5433 5434 // Call the stub. 5435 address stubAddr = StubRoutines::updateBytesCRC32(); 5436 const char *stubName = "updateBytesCRC32"; 5437 5438 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5439 stubAddr, stubName, TypePtr::BOTTOM, 5440 crc, src_start, length); 5441 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5442 set_result(result); 5443 return true; 5444 } 5445 5446 /** 5447 * Calculate CRC32 for ByteBuffer. 5448 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 5449 */ 5450 bool LibraryCallKit::inline_updateByteBufferCRC32() { 5451 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5452 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5453 // no receiver since it is static method 5454 Node* crc = argument(0); // type: int 5455 Node* src = argument(1); // type: long 5456 Node* offset = argument(3); // type: int 5457 Node* length = argument(4); // type: int 5458 5459 src = ConvL2X(src); // adjust Java long to machine word 5460 Node* base = _gvn.transform(new CastX2PNode(src)); 5461 offset = ConvI2X(offset); 5462 5463 // 'src_start' points to src array + scaled offset 5464 Node* src_start = basic_plus_adr(top(), base, offset); 5465 5466 // Call the stub. 5467 address stubAddr = StubRoutines::updateBytesCRC32(); 5468 const char *stubName = "updateBytesCRC32"; 5469 5470 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5471 stubAddr, stubName, TypePtr::BOTTOM, 5472 crc, src_start, length); 5473 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5474 set_result(result); 5475 return true; 5476 } 5477 5478 //------------------------------get_table_from_crc32c_class----------------------- 5479 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 5480 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class); 5481 assert (table != NULL, "wrong version of java.util.zip.CRC32C"); 5482 5483 return table; 5484 } 5485 5486 //------------------------------inline_updateBytesCRC32C----------------------- 5487 // 5488 // Calculate CRC32C for byte[] array. 5489 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 5490 // 5491 bool LibraryCallKit::inline_updateBytesCRC32C() { 5492 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5493 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5494 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5495 // no receiver since it is a static method 5496 Node* crc = argument(0); // type: int 5497 Node* src = argument(1); // type: oop 5498 Node* offset = argument(2); // type: int 5499 Node* end = argument(3); // type: int 5500 5501 Node* length = _gvn.transform(new SubINode(end, offset)); 5502 5503 const Type* src_type = src->Value(&_gvn); 5504 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5505 if (top_src == NULL || top_src->klass() == NULL) { 5506 // failed array check 5507 return false; 5508 } 5509 5510 // Figure out the size and type of the elements we will be copying. 5511 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5512 if (src_elem != T_BYTE) { 5513 return false; 5514 } 5515 5516 // 'src_start' points to src array + scaled offset 5517 src = must_be_not_null(src, true); 5518 Node* src_start = array_element_address(src, offset, src_elem); 5519 5520 // static final int[] byteTable in class CRC32C 5521 Node* table = get_table_from_crc32c_class(callee()->holder()); 5522 table = must_be_not_null(table, true); 5523 Node* table_start = array_element_address(table, intcon(0), T_INT); 5524 5525 // We assume that range check is done by caller. 5526 // TODO: generate range check (offset+length < src.length) in debug VM. 5527 5528 // Call the stub. 5529 address stubAddr = StubRoutines::updateBytesCRC32C(); 5530 const char *stubName = "updateBytesCRC32C"; 5531 5532 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5533 stubAddr, stubName, TypePtr::BOTTOM, 5534 crc, src_start, length, table_start); 5535 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5536 set_result(result); 5537 return true; 5538 } 5539 5540 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 5541 // 5542 // Calculate CRC32C for DirectByteBuffer. 5543 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 5544 // 5545 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 5546 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5547 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 5548 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5549 // no receiver since it is a static method 5550 Node* crc = argument(0); // type: int 5551 Node* src = argument(1); // type: long 5552 Node* offset = argument(3); // type: int 5553 Node* end = argument(4); // type: int 5554 5555 Node* length = _gvn.transform(new SubINode(end, offset)); 5556 5557 src = ConvL2X(src); // adjust Java long to machine word 5558 Node* base = _gvn.transform(new CastX2PNode(src)); 5559 offset = ConvI2X(offset); 5560 5561 // 'src_start' points to src array + scaled offset 5562 Node* src_start = basic_plus_adr(top(), base, offset); 5563 5564 // static final int[] byteTable in class CRC32C 5565 Node* table = get_table_from_crc32c_class(callee()->holder()); 5566 table = must_be_not_null(table, true); 5567 Node* table_start = array_element_address(table, intcon(0), T_INT); 5568 5569 // Call the stub. 5570 address stubAddr = StubRoutines::updateBytesCRC32C(); 5571 const char *stubName = "updateBytesCRC32C"; 5572 5573 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5574 stubAddr, stubName, TypePtr::BOTTOM, 5575 crc, src_start, length, table_start); 5576 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5577 set_result(result); 5578 return true; 5579 } 5580 5581 //------------------------------inline_updateBytesAdler32---------------------- 5582 // 5583 // Calculate Adler32 checksum for byte[] array. 5584 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 5585 // 5586 bool LibraryCallKit::inline_updateBytesAdler32() { 5587 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5588 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5589 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5590 // no receiver since it is static method 5591 Node* crc = argument(0); // type: int 5592 Node* src = argument(1); // type: oop 5593 Node* offset = argument(2); // type: int 5594 Node* length = argument(3); // type: int 5595 5596 const Type* src_type = src->Value(&_gvn); 5597 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5598 if (top_src == NULL || top_src->klass() == NULL) { 5599 // failed array check 5600 return false; 5601 } 5602 5603 // Figure out the size and type of the elements we will be copying. 5604 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5605 if (src_elem != T_BYTE) { 5606 return false; 5607 } 5608 5609 // 'src_start' points to src array + scaled offset 5610 Node* src_start = array_element_address(src, offset, src_elem); 5611 5612 // We assume that range check is done by caller. 5613 // TODO: generate range check (offset+length < src.length) in debug VM. 5614 5615 // Call the stub. 5616 address stubAddr = StubRoutines::updateBytesAdler32(); 5617 const char *stubName = "updateBytesAdler32"; 5618 5619 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5620 stubAddr, stubName, TypePtr::BOTTOM, 5621 crc, src_start, length); 5622 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5623 set_result(result); 5624 return true; 5625 } 5626 5627 //------------------------------inline_updateByteBufferAdler32--------------- 5628 // 5629 // Calculate Adler32 checksum for DirectByteBuffer. 5630 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 5631 // 5632 bool LibraryCallKit::inline_updateByteBufferAdler32() { 5633 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5634 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5635 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5636 // no receiver since it is static method 5637 Node* crc = argument(0); // type: int 5638 Node* src = argument(1); // type: long 5639 Node* offset = argument(3); // type: int 5640 Node* length = argument(4); // type: int 5641 5642 src = ConvL2X(src); // adjust Java long to machine word 5643 Node* base = _gvn.transform(new CastX2PNode(src)); 5644 offset = ConvI2X(offset); 5645 5646 // 'src_start' points to src array + scaled offset 5647 Node* src_start = basic_plus_adr(top(), base, offset); 5648 5649 // Call the stub. 5650 address stubAddr = StubRoutines::updateBytesAdler32(); 5651 const char *stubName = "updateBytesAdler32"; 5652 5653 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5654 stubAddr, stubName, TypePtr::BOTTOM, 5655 crc, src_start, length); 5656 5657 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5658 set_result(result); 5659 return true; 5660 } 5661 5662 //----------------------------inline_reference_get---------------------------- 5663 // public T java.lang.ref.Reference.get(); 5664 bool LibraryCallKit::inline_reference_get() { 5665 const int referent_offset = java_lang_ref_Reference::referent_offset; 5666 guarantee(referent_offset > 0, "should have already been set"); 5667 5668 // Get the argument: 5669 Node* reference_obj = null_check_receiver(); 5670 if (stopped()) return true; 5671 5672 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr(); 5673 assert(tinst != NULL, "obj is null"); 5674 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5675 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass(); 5676 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"), 5677 ciSymbol::make("Ljava/lang/Object;"), 5678 false); 5679 assert (field != NULL, "undefined field"); 5680 5681 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset); 5682 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 5683 5684 ciInstanceKlass* klass = env()->Object_klass(); 5685 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass); 5686 5687 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 5688 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators); 5689 // Add memory barrier to prevent commoning reads from this field 5690 // across safepoint since GC can change its value. 5691 insert_mem_bar(Op_MemBarCPUOrder); 5692 5693 set_result(result); 5694 return true; 5695 } 5696 5697 5698 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 5699 bool is_exact=true, bool is_static=false, 5700 ciInstanceKlass * fromKls=NULL) { 5701 if (fromKls == NULL) { 5702 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 5703 assert(tinst != NULL, "obj is null"); 5704 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5705 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 5706 fromKls = tinst->klass()->as_instance_klass(); 5707 } else { 5708 assert(is_static, "only for static field access"); 5709 } 5710 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 5711 ciSymbol::make(fieldTypeString), 5712 is_static); 5713 5714 assert (field != NULL, "undefined field"); 5715 if (field == NULL) return (Node *) NULL; 5716 5717 if (is_static) { 5718 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 5719 fromObj = makecon(tip); 5720 } 5721 5722 // Next code copied from Parse::do_get_xxx(): 5723 5724 // Compute address and memory type. 5725 int offset = field->offset_in_bytes(); 5726 bool is_vol = field->is_volatile(); 5727 ciType* field_klass = field->type(); 5728 assert(field_klass->is_loaded(), "should be loaded"); 5729 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 5730 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 5731 BasicType bt = field->layout_type(); 5732 5733 // Build the resultant type of the load 5734 const Type *type; 5735 if (bt == T_OBJECT) { 5736 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 5737 } else { 5738 type = Type::get_const_basic_type(bt); 5739 } 5740 5741 DecoratorSet decorators = IN_HEAP; 5742 5743 if (is_vol) { 5744 decorators |= MO_SEQ_CST; 5745 } 5746 5747 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 5748 } 5749 5750 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 5751 bool is_exact = true, bool is_static = false, 5752 ciInstanceKlass * fromKls = NULL) { 5753 if (fromKls == NULL) { 5754 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 5755 assert(tinst != NULL, "obj is null"); 5756 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5757 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 5758 fromKls = tinst->klass()->as_instance_klass(); 5759 } 5760 else { 5761 assert(is_static, "only for static field access"); 5762 } 5763 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 5764 ciSymbol::make(fieldTypeString), 5765 is_static); 5766 5767 assert(field != NULL, "undefined field"); 5768 assert(!field->is_volatile(), "not defined for volatile fields"); 5769 5770 if (is_static) { 5771 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 5772 fromObj = makecon(tip); 5773 } 5774 5775 // Next code copied from Parse::do_get_xxx(): 5776 5777 // Compute address and memory type. 5778 int offset = field->offset_in_bytes(); 5779 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 5780 5781 return adr; 5782 } 5783 5784 //------------------------------inline_aescrypt_Block----------------------- 5785 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 5786 address stubAddr = NULL; 5787 const char *stubName; 5788 assert(UseAES, "need AES instruction support"); 5789 5790 switch(id) { 5791 case vmIntrinsics::_aescrypt_encryptBlock: 5792 stubAddr = StubRoutines::aescrypt_encryptBlock(); 5793 stubName = "aescrypt_encryptBlock"; 5794 break; 5795 case vmIntrinsics::_aescrypt_decryptBlock: 5796 stubAddr = StubRoutines::aescrypt_decryptBlock(); 5797 stubName = "aescrypt_decryptBlock"; 5798 break; 5799 default: 5800 break; 5801 } 5802 if (stubAddr == NULL) return false; 5803 5804 Node* aescrypt_object = argument(0); 5805 Node* src = argument(1); 5806 Node* src_offset = argument(2); 5807 Node* dest = argument(3); 5808 Node* dest_offset = argument(4); 5809 5810 src = must_be_not_null(src, true); 5811 dest = must_be_not_null(dest, true); 5812 5813 // (1) src and dest are arrays. 5814 const Type* src_type = src->Value(&_gvn); 5815 const Type* dest_type = dest->Value(&_gvn); 5816 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5817 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5818 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 5819 5820 // for the quick and dirty code we will skip all the checks. 5821 // we are just trying to get the call to be generated. 5822 Node* src_start = src; 5823 Node* dest_start = dest; 5824 if (src_offset != NULL || dest_offset != NULL) { 5825 assert(src_offset != NULL && dest_offset != NULL, ""); 5826 src_start = array_element_address(src, src_offset, T_BYTE); 5827 dest_start = array_element_address(dest, dest_offset, T_BYTE); 5828 } 5829 5830 // now need to get the start of its expanded key array 5831 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 5832 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 5833 if (k_start == NULL) return false; 5834 5835 if (Matcher::pass_original_key_for_aes()) { 5836 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 5837 // compatibility issues between Java key expansion and SPARC crypto instructions 5838 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 5839 if (original_k_start == NULL) return false; 5840 5841 // Call the stub. 5842 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 5843 stubAddr, stubName, TypePtr::BOTTOM, 5844 src_start, dest_start, k_start, original_k_start); 5845 } else { 5846 // Call the stub. 5847 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 5848 stubAddr, stubName, TypePtr::BOTTOM, 5849 src_start, dest_start, k_start); 5850 } 5851 5852 return true; 5853 } 5854 5855 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 5856 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 5857 address stubAddr = NULL; 5858 const char *stubName = NULL; 5859 5860 assert(UseAES, "need AES instruction support"); 5861 5862 switch(id) { 5863 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 5864 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 5865 stubName = "cipherBlockChaining_encryptAESCrypt"; 5866 break; 5867 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 5868 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 5869 stubName = "cipherBlockChaining_decryptAESCrypt"; 5870 break; 5871 default: 5872 break; 5873 } 5874 if (stubAddr == NULL) return false; 5875 5876 Node* cipherBlockChaining_object = argument(0); 5877 Node* src = argument(1); 5878 Node* src_offset = argument(2); 5879 Node* len = argument(3); 5880 Node* dest = argument(4); 5881 Node* dest_offset = argument(5); 5882 5883 src = must_be_not_null(src, false); 5884 dest = must_be_not_null(dest, false); 5885 5886 // (1) src and dest are arrays. 5887 const Type* src_type = src->Value(&_gvn); 5888 const Type* dest_type = dest->Value(&_gvn); 5889 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5890 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5891 assert (top_src != NULL && top_src->klass() != NULL 5892 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 5893 5894 // checks are the responsibility of the caller 5895 Node* src_start = src; 5896 Node* dest_start = dest; 5897 if (src_offset != NULL || dest_offset != NULL) { 5898 assert(src_offset != NULL && dest_offset != NULL, ""); 5899 src_start = array_element_address(src, src_offset, T_BYTE); 5900 dest_start = array_element_address(dest, dest_offset, T_BYTE); 5901 } 5902 5903 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 5904 // (because of the predicated logic executed earlier). 5905 // so we cast it here safely. 5906 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 5907 5908 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 5909 if (embeddedCipherObj == NULL) return false; 5910 5911 // cast it to what we know it will be at runtime 5912 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 5913 assert(tinst != NULL, "CBC obj is null"); 5914 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded"); 5915 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 5916 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 5917 5918 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 5919 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 5920 const TypeOopPtr* xtype = aklass->as_instance_type(); 5921 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 5922 aescrypt_object = _gvn.transform(aescrypt_object); 5923 5924 // we need to get the start of the aescrypt_object's expanded key array 5925 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 5926 if (k_start == NULL) return false; 5927 5928 // similarly, get the start address of the r vector 5929 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false); 5930 if (objRvec == NULL) return false; 5931 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 5932 5933 Node* cbcCrypt; 5934 if (Matcher::pass_original_key_for_aes()) { 5935 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 5936 // compatibility issues between Java key expansion and SPARC crypto instructions 5937 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 5938 if (original_k_start == NULL) return false; 5939 5940 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start 5941 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 5942 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 5943 stubAddr, stubName, TypePtr::BOTTOM, 5944 src_start, dest_start, k_start, r_start, len, original_k_start); 5945 } else { 5946 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 5947 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 5948 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 5949 stubAddr, stubName, TypePtr::BOTTOM, 5950 src_start, dest_start, k_start, r_start, len); 5951 } 5952 5953 // return cipher length (int) 5954 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 5955 set_result(retvalue); 5956 return true; 5957 } 5958 5959 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 5960 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 5961 address stubAddr = NULL; 5962 const char *stubName = NULL; 5963 5964 assert(UseAES, "need AES instruction support"); 5965 5966 switch (id) { 5967 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 5968 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 5969 stubName = "electronicCodeBook_encryptAESCrypt"; 5970 break; 5971 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 5972 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 5973 stubName = "electronicCodeBook_decryptAESCrypt"; 5974 break; 5975 default: 5976 break; 5977 } 5978 5979 if (stubAddr == NULL) return false; 5980 5981 Node* electronicCodeBook_object = argument(0); 5982 Node* src = argument(1); 5983 Node* src_offset = argument(2); 5984 Node* len = argument(3); 5985 Node* dest = argument(4); 5986 Node* dest_offset = argument(5); 5987 5988 // (1) src and dest are arrays. 5989 const Type* src_type = src->Value(&_gvn); 5990 const Type* dest_type = dest->Value(&_gvn); 5991 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5992 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5993 assert(top_src != NULL && top_src->klass() != NULL 5994 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 5995 5996 // checks are the responsibility of the caller 5997 Node* src_start = src; 5998 Node* dest_start = dest; 5999 if (src_offset != NULL || dest_offset != NULL) { 6000 assert(src_offset != NULL && dest_offset != NULL, ""); 6001 src_start = array_element_address(src, src_offset, T_BYTE); 6002 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6003 } 6004 6005 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6006 // (because of the predicated logic executed earlier). 6007 // so we cast it here safely. 6008 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6009 6010 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6011 if (embeddedCipherObj == NULL) return false; 6012 6013 // cast it to what we know it will be at runtime 6014 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 6015 assert(tinst != NULL, "ECB obj is null"); 6016 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded"); 6017 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6018 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6019 6020 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6021 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6022 const TypeOopPtr* xtype = aklass->as_instance_type(); 6023 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6024 aescrypt_object = _gvn.transform(aescrypt_object); 6025 6026 // we need to get the start of the aescrypt_object's expanded key array 6027 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6028 if (k_start == NULL) return false; 6029 6030 Node* ecbCrypt; 6031 if (Matcher::pass_original_key_for_aes()) { 6032 // no SPARC version for AES/ECB intrinsics now. 6033 return false; 6034 } 6035 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6036 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 6037 OptoRuntime::electronicCodeBook_aescrypt_Type(), 6038 stubAddr, stubName, TypePtr::BOTTOM, 6039 src_start, dest_start, k_start, len); 6040 6041 // return cipher length (int) 6042 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 6043 set_result(retvalue); 6044 return true; 6045 } 6046 6047 //------------------------------inline_counterMode_AESCrypt----------------------- 6048 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 6049 assert(UseAES, "need AES instruction support"); 6050 if (!UseAESCTRIntrinsics) return false; 6051 6052 address stubAddr = NULL; 6053 const char *stubName = NULL; 6054 if (id == vmIntrinsics::_counterMode_AESCrypt) { 6055 stubAddr = StubRoutines::counterMode_AESCrypt(); 6056 stubName = "counterMode_AESCrypt"; 6057 } 6058 if (stubAddr == NULL) return false; 6059 6060 Node* counterMode_object = argument(0); 6061 Node* src = argument(1); 6062 Node* src_offset = argument(2); 6063 Node* len = argument(3); 6064 Node* dest = argument(4); 6065 Node* dest_offset = argument(5); 6066 6067 // (1) src and dest are arrays. 6068 const Type* src_type = src->Value(&_gvn); 6069 const Type* dest_type = dest->Value(&_gvn); 6070 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6071 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6072 assert(top_src != NULL && top_src->klass() != NULL && 6073 top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6074 6075 // checks are the responsibility of the caller 6076 Node* src_start = src; 6077 Node* dest_start = dest; 6078 if (src_offset != NULL || dest_offset != NULL) { 6079 assert(src_offset != NULL && dest_offset != NULL, ""); 6080 src_start = array_element_address(src, src_offset, T_BYTE); 6081 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6082 } 6083 6084 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6085 // (because of the predicated logic executed earlier). 6086 // so we cast it here safely. 6087 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6088 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6089 if (embeddedCipherObj == NULL) return false; 6090 // cast it to what we know it will be at runtime 6091 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 6092 assert(tinst != NULL, "CTR obj is null"); 6093 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded"); 6094 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6095 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6096 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6097 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6098 const TypeOopPtr* xtype = aklass->as_instance_type(); 6099 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6100 aescrypt_object = _gvn.transform(aescrypt_object); 6101 // we need to get the start of the aescrypt_object's expanded key array 6102 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6103 if (k_start == NULL) return false; 6104 // similarly, get the start address of the r vector 6105 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false); 6106 if (obj_counter == NULL) return false; 6107 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 6108 6109 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false); 6110 if (saved_encCounter == NULL) return false; 6111 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 6112 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 6113 6114 Node* ctrCrypt; 6115 if (Matcher::pass_original_key_for_aes()) { 6116 // no SPARC version for AES/CTR intrinsics now. 6117 return false; 6118 } 6119 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6120 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6121 OptoRuntime::counterMode_aescrypt_Type(), 6122 stubAddr, stubName, TypePtr::BOTTOM, 6123 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 6124 6125 // return cipher length (int) 6126 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 6127 set_result(retvalue); 6128 return true; 6129 } 6130 6131 //------------------------------get_key_start_from_aescrypt_object----------------------- 6132 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 6133 #if defined(PPC64) || defined(S390) 6134 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 6135 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 6136 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 6137 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]). 6138 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false); 6139 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6140 if (objSessionK == NULL) { 6141 return (Node *) NULL; 6142 } 6143 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS); 6144 #else 6145 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false); 6146 #endif // PPC64 6147 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6148 if (objAESCryptKey == NULL) return (Node *) NULL; 6149 6150 // now have the array, need to get the start address of the K array 6151 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 6152 return k_start; 6153 } 6154 6155 //------------------------------get_original_key_start_from_aescrypt_object----------------------- 6156 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { 6157 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); 6158 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6159 if (objAESCryptKey == NULL) return (Node *) NULL; 6160 6161 // now have the array, need to get the start address of the lastKey array 6162 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); 6163 return original_k_start; 6164 } 6165 6166 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 6167 // Return node representing slow path of predicate check. 6168 // the pseudo code we want to emulate with this predicate is: 6169 // for encryption: 6170 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6171 // for decryption: 6172 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6173 // note cipher==plain is more conservative than the original java code but that's OK 6174 // 6175 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 6176 // The receiver was checked for NULL already. 6177 Node* objCBC = argument(0); 6178 6179 Node* src = argument(1); 6180 Node* dest = argument(4); 6181 6182 // Load embeddedCipher field of CipherBlockChaining object. 6183 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6184 6185 // get AESCrypt klass for instanceOf check 6186 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6187 // will have same classloader as CipherBlockChaining object 6188 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 6189 assert(tinst != NULL, "CBCobj is null"); 6190 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded"); 6191 6192 // we want to do an instanceof comparison against the AESCrypt class 6193 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6194 if (!klass_AESCrypt->is_loaded()) { 6195 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6196 Node* ctrl = control(); 6197 set_control(top()); // no regular fast path 6198 return ctrl; 6199 } 6200 6201 src = must_be_not_null(src, true); 6202 dest = must_be_not_null(dest, true); 6203 6204 // Resolve oops to stable for CmpP below. 6205 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6206 6207 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6208 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6209 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6210 6211 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6212 6213 // for encryption, we are done 6214 if (!decrypting) 6215 return instof_false; // even if it is NULL 6216 6217 // for decryption, we need to add a further check to avoid 6218 // taking the intrinsic path when cipher and plain are the same 6219 // see the original java code for why. 6220 RegionNode* region = new RegionNode(3); 6221 region->init_req(1, instof_false); 6222 6223 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6224 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6225 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6226 region->init_req(2, src_dest_conjoint); 6227 6228 record_for_igvn(region); 6229 return _gvn.transform(region); 6230 } 6231 6232 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 6233 // Return node representing slow path of predicate check. 6234 // the pseudo code we want to emulate with this predicate is: 6235 // for encryption: 6236 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6237 // for decryption: 6238 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6239 // note cipher==plain is more conservative than the original java code but that's OK 6240 // 6241 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 6242 // The receiver was checked for NULL already. 6243 Node* objECB = argument(0); 6244 6245 // Load embeddedCipher field of ElectronicCodeBook object. 6246 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6247 6248 // get AESCrypt klass for instanceOf check 6249 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6250 // will have same classloader as ElectronicCodeBook object 6251 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 6252 assert(tinst != NULL, "ECBobj is null"); 6253 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded"); 6254 6255 // we want to do an instanceof comparison against the AESCrypt class 6256 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6257 if (!klass_AESCrypt->is_loaded()) { 6258 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6259 Node* ctrl = control(); 6260 set_control(top()); // no regular fast path 6261 return ctrl; 6262 } 6263 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6264 6265 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6266 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6267 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6268 6269 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6270 6271 // for encryption, we are done 6272 if (!decrypting) 6273 return instof_false; // even if it is NULL 6274 6275 // for decryption, we need to add a further check to avoid 6276 // taking the intrinsic path when cipher and plain are the same 6277 // see the original java code for why. 6278 RegionNode* region = new RegionNode(3); 6279 region->init_req(1, instof_false); 6280 Node* src = argument(1); 6281 Node* dest = argument(4); 6282 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6283 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6284 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6285 region->init_req(2, src_dest_conjoint); 6286 6287 record_for_igvn(region); 6288 return _gvn.transform(region); 6289 } 6290 6291 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 6292 // Return node representing slow path of predicate check. 6293 // the pseudo code we want to emulate with this predicate is: 6294 // for encryption: 6295 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6296 // for decryption: 6297 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6298 // note cipher==plain is more conservative than the original java code but that's OK 6299 // 6300 6301 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 6302 // The receiver was checked for NULL already. 6303 Node* objCTR = argument(0); 6304 6305 // Load embeddedCipher field of CipherBlockChaining object. 6306 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6307 6308 // get AESCrypt klass for instanceOf check 6309 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6310 // will have same classloader as CipherBlockChaining object 6311 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 6312 assert(tinst != NULL, "CTRobj is null"); 6313 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded"); 6314 6315 // we want to do an instanceof comparison against the AESCrypt class 6316 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6317 if (!klass_AESCrypt->is_loaded()) { 6318 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6319 Node* ctrl = control(); 6320 set_control(top()); // no regular fast path 6321 return ctrl; 6322 } 6323 6324 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6325 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6326 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6327 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6328 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6329 6330 return instof_false; // even if it is NULL 6331 } 6332 6333 //------------------------------inline_ghash_processBlocks 6334 bool LibraryCallKit::inline_ghash_processBlocks() { 6335 address stubAddr; 6336 const char *stubName; 6337 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 6338 6339 stubAddr = StubRoutines::ghash_processBlocks(); 6340 stubName = "ghash_processBlocks"; 6341 6342 Node* data = argument(0); 6343 Node* offset = argument(1); 6344 Node* len = argument(2); 6345 Node* state = argument(3); 6346 Node* subkeyH = argument(4); 6347 6348 state = must_be_not_null(state, true); 6349 subkeyH = must_be_not_null(subkeyH, true); 6350 data = must_be_not_null(data, true); 6351 6352 Node* state_start = array_element_address(state, intcon(0), T_LONG); 6353 assert(state_start, "state is NULL"); 6354 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 6355 assert(subkeyH_start, "subkeyH is NULL"); 6356 Node* data_start = array_element_address(data, offset, T_BYTE); 6357 assert(data_start, "data is NULL"); 6358 6359 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 6360 OptoRuntime::ghash_processBlocks_Type(), 6361 stubAddr, stubName, TypePtr::BOTTOM, 6362 state_start, subkeyH_start, data_start, len); 6363 return true; 6364 } 6365 6366 bool LibraryCallKit::inline_base64_encodeBlock() { 6367 address stubAddr; 6368 const char *stubName; 6369 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 6370 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 6371 stubAddr = StubRoutines::base64_encodeBlock(); 6372 stubName = "encodeBlock"; 6373 6374 if (!stubAddr) return false; 6375 Node* base64obj = argument(0); 6376 Node* src = argument(1); 6377 Node* offset = argument(2); 6378 Node* len = argument(3); 6379 Node* dest = argument(4); 6380 Node* dp = argument(5); 6381 Node* isURL = argument(6); 6382 6383 src = must_be_not_null(src, true); 6384 dest = must_be_not_null(dest, true); 6385 6386 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 6387 assert(src_start, "source array is NULL"); 6388 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 6389 assert(dest_start, "destination array is NULL"); 6390 6391 Node* base64 = make_runtime_call(RC_LEAF, 6392 OptoRuntime::base64_encodeBlock_Type(), 6393 stubAddr, stubName, TypePtr::BOTTOM, 6394 src_start, offset, len, dest_start, dp, isURL); 6395 return true; 6396 } 6397 6398 //------------------------------inline_sha_implCompress----------------------- 6399 // 6400 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 6401 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 6402 // 6403 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 6404 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 6405 // 6406 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 6407 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 6408 // 6409 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) { 6410 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 6411 6412 Node* sha_obj = argument(0); 6413 Node* src = argument(1); // type oop 6414 Node* ofs = argument(2); // type int 6415 6416 const Type* src_type = src->Value(&_gvn); 6417 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6418 if (top_src == NULL || top_src->klass() == NULL) { 6419 // failed array check 6420 return false; 6421 } 6422 // Figure out the size and type of the elements we will be copying. 6423 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6424 if (src_elem != T_BYTE) { 6425 return false; 6426 } 6427 // 'src_start' points to src array + offset 6428 src = must_be_not_null(src, true); 6429 Node* src_start = array_element_address(src, ofs, src_elem); 6430 Node* state = NULL; 6431 address stubAddr; 6432 const char *stubName; 6433 6434 switch(id) { 6435 case vmIntrinsics::_sha_implCompress: 6436 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 6437 state = get_state_from_sha_object(sha_obj); 6438 stubAddr = StubRoutines::sha1_implCompress(); 6439 stubName = "sha1_implCompress"; 6440 break; 6441 case vmIntrinsics::_sha2_implCompress: 6442 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 6443 state = get_state_from_sha_object(sha_obj); 6444 stubAddr = StubRoutines::sha256_implCompress(); 6445 stubName = "sha256_implCompress"; 6446 break; 6447 case vmIntrinsics::_sha5_implCompress: 6448 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 6449 state = get_state_from_sha5_object(sha_obj); 6450 stubAddr = StubRoutines::sha512_implCompress(); 6451 stubName = "sha512_implCompress"; 6452 break; 6453 default: 6454 fatal_unexpected_iid(id); 6455 return false; 6456 } 6457 if (state == NULL) return false; 6458 6459 assert(stubAddr != NULL, "Stub is generated"); 6460 if (stubAddr == NULL) return false; 6461 6462 // Call the stub. 6463 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(), 6464 stubAddr, stubName, TypePtr::BOTTOM, 6465 src_start, state); 6466 6467 return true; 6468 } 6469 6470 //------------------------------inline_digestBase_implCompressMB----------------------- 6471 // 6472 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array. 6473 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 6474 // 6475 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 6476 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6477 "need SHA1/SHA256/SHA512 instruction support"); 6478 assert((uint)predicate < 3, "sanity"); 6479 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 6480 6481 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already. 6482 Node* src = argument(1); // byte[] array 6483 Node* ofs = argument(2); // type int 6484 Node* limit = argument(3); // type int 6485 6486 const Type* src_type = src->Value(&_gvn); 6487 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6488 if (top_src == NULL || top_src->klass() == NULL) { 6489 // failed array check 6490 return false; 6491 } 6492 // Figure out the size and type of the elements we will be copying. 6493 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6494 if (src_elem != T_BYTE) { 6495 return false; 6496 } 6497 // 'src_start' points to src array + offset 6498 src = must_be_not_null(src, false); 6499 Node* src_start = array_element_address(src, ofs, src_elem); 6500 6501 const char* klass_SHA_name = NULL; 6502 const char* stub_name = NULL; 6503 address stub_addr = NULL; 6504 bool long_state = false; 6505 6506 switch (predicate) { 6507 case 0: 6508 if (UseSHA1Intrinsics) { 6509 klass_SHA_name = "sun/security/provider/SHA"; 6510 stub_name = "sha1_implCompressMB"; 6511 stub_addr = StubRoutines::sha1_implCompressMB(); 6512 } 6513 break; 6514 case 1: 6515 if (UseSHA256Intrinsics) { 6516 klass_SHA_name = "sun/security/provider/SHA2"; 6517 stub_name = "sha256_implCompressMB"; 6518 stub_addr = StubRoutines::sha256_implCompressMB(); 6519 } 6520 break; 6521 case 2: 6522 if (UseSHA512Intrinsics) { 6523 klass_SHA_name = "sun/security/provider/SHA5"; 6524 stub_name = "sha512_implCompressMB"; 6525 stub_addr = StubRoutines::sha512_implCompressMB(); 6526 long_state = true; 6527 } 6528 break; 6529 default: 6530 fatal("unknown SHA intrinsic predicate: %d", predicate); 6531 } 6532 if (klass_SHA_name != NULL) { 6533 assert(stub_addr != NULL, "Stub is generated"); 6534 if (stub_addr == NULL) return false; 6535 6536 // get DigestBase klass to lookup for SHA klass 6537 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 6538 assert(tinst != NULL, "digestBase_obj is not instance???"); 6539 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6540 6541 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 6542 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded"); 6543 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 6544 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit); 6545 } 6546 return false; 6547 } 6548 //------------------------------inline_sha_implCompressMB----------------------- 6549 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA, 6550 bool long_state, address stubAddr, const char *stubName, 6551 Node* src_start, Node* ofs, Node* limit) { 6552 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA); 6553 const TypeOopPtr* xtype = aklass->as_instance_type(); 6554 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 6555 sha_obj = _gvn.transform(sha_obj); 6556 6557 Node* state; 6558 if (long_state) { 6559 state = get_state_from_sha5_object(sha_obj); 6560 } else { 6561 state = get_state_from_sha_object(sha_obj); 6562 } 6563 if (state == NULL) return false; 6564 6565 // Call the stub. 6566 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6567 OptoRuntime::digestBase_implCompressMB_Type(), 6568 stubAddr, stubName, TypePtr::BOTTOM, 6569 src_start, state, ofs, limit); 6570 // return ofs (int) 6571 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6572 set_result(result); 6573 6574 return true; 6575 } 6576 6577 //------------------------------get_state_from_sha_object----------------------- 6578 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) { 6579 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false); 6580 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2"); 6581 if (sha_state == NULL) return (Node *) NULL; 6582 6583 // now have the array, need to get the start address of the state array 6584 Node* state = array_element_address(sha_state, intcon(0), T_INT); 6585 return state; 6586 } 6587 6588 //------------------------------get_state_from_sha5_object----------------------- 6589 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) { 6590 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false); 6591 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5"); 6592 if (sha_state == NULL) return (Node *) NULL; 6593 6594 // now have the array, need to get the start address of the state array 6595 Node* state = array_element_address(sha_state, intcon(0), T_LONG); 6596 return state; 6597 } 6598 6599 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 6600 // Return node representing slow path of predicate check. 6601 // the pseudo code we want to emulate with this predicate is: 6602 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath 6603 // 6604 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 6605 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6606 "need SHA1/SHA256/SHA512 instruction support"); 6607 assert((uint)predicate < 3, "sanity"); 6608 6609 // The receiver was checked for NULL already. 6610 Node* digestBaseObj = argument(0); 6611 6612 // get DigestBase klass for instanceOf check 6613 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 6614 assert(tinst != NULL, "digestBaseObj is null"); 6615 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6616 6617 const char* klass_SHA_name = NULL; 6618 switch (predicate) { 6619 case 0: 6620 if (UseSHA1Intrinsics) { 6621 // we want to do an instanceof comparison against the SHA class 6622 klass_SHA_name = "sun/security/provider/SHA"; 6623 } 6624 break; 6625 case 1: 6626 if (UseSHA256Intrinsics) { 6627 // we want to do an instanceof comparison against the SHA2 class 6628 klass_SHA_name = "sun/security/provider/SHA2"; 6629 } 6630 break; 6631 case 2: 6632 if (UseSHA512Intrinsics) { 6633 // we want to do an instanceof comparison against the SHA5 class 6634 klass_SHA_name = "sun/security/provider/SHA5"; 6635 } 6636 break; 6637 default: 6638 fatal("unknown SHA intrinsic predicate: %d", predicate); 6639 } 6640 6641 ciKlass* klass_SHA = NULL; 6642 if (klass_SHA_name != NULL) { 6643 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 6644 } 6645 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) { 6646 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 6647 Node* ctrl = control(); 6648 set_control(top()); // no intrinsic path 6649 return ctrl; 6650 } 6651 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 6652 6653 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA))); 6654 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1))); 6655 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6656 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6657 6658 return instof_false; // even if it is NULL 6659 } 6660 6661 //-------------inline_fma----------------------------------- 6662 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 6663 Node *a = NULL; 6664 Node *b = NULL; 6665 Node *c = NULL; 6666 Node* result = NULL; 6667 switch (id) { 6668 case vmIntrinsics::_fmaD: 6669 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 6670 // no receiver since it is static method 6671 a = round_double_node(argument(0)); 6672 b = round_double_node(argument(2)); 6673 c = round_double_node(argument(4)); 6674 result = _gvn.transform(new FmaDNode(control(), a, b, c)); 6675 break; 6676 case vmIntrinsics::_fmaF: 6677 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 6678 a = argument(0); 6679 b = argument(1); 6680 c = argument(2); 6681 result = _gvn.transform(new FmaFNode(control(), a, b, c)); 6682 break; 6683 default: 6684 fatal_unexpected_iid(id); break; 6685 } 6686 set_result(result); 6687 return true; 6688 } 6689 6690 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 6691 // argument(0) is receiver 6692 Node* codePoint = argument(1); 6693 Node* n = NULL; 6694 6695 switch (id) { 6696 case vmIntrinsics::_isDigit : 6697 n = new DigitNode(control(), codePoint); 6698 break; 6699 case vmIntrinsics::_isLowerCase : 6700 n = new LowerCaseNode(control(), codePoint); 6701 break; 6702 case vmIntrinsics::_isUpperCase : 6703 n = new UpperCaseNode(control(), codePoint); 6704 break; 6705 case vmIntrinsics::_isWhitespace : 6706 n = new WhitespaceNode(control(), codePoint); 6707 break; 6708 default: 6709 fatal_unexpected_iid(id); 6710 } 6711 6712 set_result(_gvn.transform(n)); 6713 return true; 6714 } 6715 6716 //------------------------------inline_fp_min_max------------------------------ 6717 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) { 6718 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416. 6719 6720 // The intrinsic should be used only when the API branches aren't predictable, 6721 // the last one performing the most important comparison. The following heuristic 6722 // uses the branch statistics to eventually bail out if necessary. 6723 6724 ciMethodData *md = callee()->method_data(); 6725 6726 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) { 6727 ciCallProfile cp = caller()->call_profile_at_bci(bci()); 6728 6729 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) { 6730 // Bail out if the call-site didn't contribute enough to the statistics. 6731 return false; 6732 } 6733 6734 uint taken = 0, not_taken = 0; 6735 6736 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) { 6737 if (p->is_BranchData()) { 6738 taken = ((ciBranchData*)p)->taken(); 6739 not_taken = ((ciBranchData*)p)->not_taken(); 6740 } 6741 } 6742 6743 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count()); 6744 balance = balance < 0 ? -balance : balance; 6745 if ( balance > 0.2 ) { 6746 // Bail out if the most important branch is predictable enough. 6747 return false; 6748 } 6749 } 6750 */ 6751 6752 Node *a = NULL; 6753 Node *b = NULL; 6754 Node *n = NULL; 6755 switch (id) { 6756 case vmIntrinsics::_maxF: 6757 case vmIntrinsics::_minF: 6758 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 6759 a = argument(0); 6760 b = argument(1); 6761 break; 6762 case vmIntrinsics::_maxD: 6763 case vmIntrinsics::_minD: 6764 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 6765 a = round_double_node(argument(0)); 6766 b = round_double_node(argument(2)); 6767 break; 6768 default: 6769 fatal_unexpected_iid(id); 6770 break; 6771 } 6772 switch (id) { 6773 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break; 6774 case vmIntrinsics::_minF: n = new MinFNode(a, b); break; 6775 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break; 6776 case vmIntrinsics::_minD: n = new MinDNode(a, b); break; 6777 default: fatal_unexpected_iid(id); break; 6778 } 6779 set_result(_gvn.transform(n)); 6780 return true; 6781 } 6782 6783 bool LibraryCallKit::inline_profileBoolean() { 6784 Node* counts = argument(1); 6785 const TypeAryPtr* ary = NULL; 6786 ciArray* aobj = NULL; 6787 if (counts->is_Con() 6788 && (ary = counts->bottom_type()->isa_aryptr()) != NULL 6789 && (aobj = ary->const_oop()->as_array()) != NULL 6790 && (aobj->length() == 2)) { 6791 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 6792 jint false_cnt = aobj->element_value(0).as_int(); 6793 jint true_cnt = aobj->element_value(1).as_int(); 6794 6795 if (C->log() != NULL) { 6796 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 6797 false_cnt, true_cnt); 6798 } 6799 6800 if (false_cnt + true_cnt == 0) { 6801 // According to profile, never executed. 6802 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6803 Deoptimization::Action_reinterpret); 6804 return true; 6805 } 6806 6807 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 6808 // is a number of each value occurrences. 6809 Node* result = argument(0); 6810 if (false_cnt == 0 || true_cnt == 0) { 6811 // According to profile, one value has been never seen. 6812 int expected_val = (false_cnt == 0) ? 1 : 0; 6813 6814 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 6815 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 6816 6817 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 6818 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 6819 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 6820 6821 { // Slow path: uncommon trap for never seen value and then reexecute 6822 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 6823 // the value has been seen at least once. 6824 PreserveJVMState pjvms(this); 6825 PreserveReexecuteState preexecs(this); 6826 jvms()->set_should_reexecute(true); 6827 6828 set_control(slow_path); 6829 set_i_o(i_o()); 6830 6831 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6832 Deoptimization::Action_reinterpret); 6833 } 6834 // The guard for never seen value enables sharpening of the result and 6835 // returning a constant. It allows to eliminate branches on the same value 6836 // later on. 6837 set_control(fast_path); 6838 result = intcon(expected_val); 6839 } 6840 // Stop profiling. 6841 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 6842 // By replacing method body with profile data (represented as ProfileBooleanNode 6843 // on IR level) we effectively disable profiling. 6844 // It enables full speed execution once optimized code is generated. 6845 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 6846 C->record_for_igvn(profile); 6847 set_result(profile); 6848 return true; 6849 } else { 6850 // Continue profiling. 6851 // Profile data isn't available at the moment. So, execute method's bytecode version. 6852 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 6853 // is compiled and counters aren't available since corresponding MethodHandle 6854 // isn't a compile-time constant. 6855 return false; 6856 } 6857 } 6858 6859 bool LibraryCallKit::inline_isCompileConstant() { 6860 Node* n = argument(0); 6861 set_result(n->is_Con() ? intcon(1) : intcon(0)); 6862 return true; 6863 }